10 More than a Pretty Picture: Visual Thinking in Network Studies
123
More than a Pretty Picture: Visual Thinking in Network Studies James Moody Duke Network Analysis Center (DNAC) https://dnac.ssri.duke.edu / Duke University http://tinyurl.com/duke-snh This work is supported by NIH grants DA12831 and HD41877. Thanks to Dan McFarland, Skye Bender-deMoll, Martina Morris, Lisa Keister, Scott Gest and Peter Mucha for discussion and comments related to this presentation. Please send any correspondence to James Moody, Department of Sociology, Duke University, Durham NC 27705. email: [email protected]
Transcript of 10 More than a Pretty Picture: Visual Thinking in Network Studies
More than a Pretty PictureVisual Thinking in Network Studies
James MoodyDuke Network Analysis Center (DNAC)
httpsdnacssridukeedu Duke University
httptinyurlcomduke-snh
This work is supported by NIH grants DA12831 and HD41877 Thanks to Dan McFarland Skye Bender-deMoll Martina Morris Lisa Keister Scott Gest and Peter Mucha for discussion and comments related to this presentation Please send any correspondence to James Moody Department of Sociology Duke University Durham NC 27705 email jmoody77socdukeedu
Introduction
We live in a connected world
ldquoTo speak of social life is to speak of the association between people ndash their associating in work and in play in love and in war to trade or to worship to help or to hinder It is in the social relations men establish that their interests find expression and their desires become realizedrdquo
Peter M BlauExchange and Power in Social Life 1964
1934 NYTime Moreno claims this work was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Introduction
We live in a connected world
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
Introduction
From the beginning network science has been a visual science rich with strange images purporting to illuminate our understanding of the world
1) Introduction2) A Little Historyhellip3) Why visualize networks at all4) What makes a good (scientific) visualization5) Network Visualization Strategies
1) Common heuristics2) Layout Features3) Static dynamic interactive
6) Guiding Idea Build theory into visualization7) Common Challenges
8) Can We Develop a visual network language9) Conclusion10) Some fun exampleshellipif timehellip(ha)
Visual Network Thinking Outline
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Introduction
We live in a connected world
ldquoTo speak of social life is to speak of the association between people ndash their associating in work and in play in love and in war to trade or to worship to help or to hinder It is in the social relations men establish that their interests find expression and their desires become realizedrdquo
Peter M BlauExchange and Power in Social Life 1964
1934 NYTime Moreno claims this work was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Introduction
We live in a connected world
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
Introduction
From the beginning network science has been a visual science rich with strange images purporting to illuminate our understanding of the world
1) Introduction2) A Little Historyhellip3) Why visualize networks at all4) What makes a good (scientific) visualization5) Network Visualization Strategies
1) Common heuristics2) Layout Features3) Static dynamic interactive
6) Guiding Idea Build theory into visualization7) Common Challenges
8) Can We Develop a visual network language9) Conclusion10) Some fun exampleshellipif timehellip(ha)
Visual Network Thinking Outline
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
1934 NYTime Moreno claims this work was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Introduction
We live in a connected world
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
Introduction
From the beginning network science has been a visual science rich with strange images purporting to illuminate our understanding of the world
1) Introduction2) A Little Historyhellip3) Why visualize networks at all4) What makes a good (scientific) visualization5) Network Visualization Strategies
1) Common heuristics2) Layout Features3) Static dynamic interactive
6) Guiding Idea Build theory into visualization7) Common Challenges
8) Can We Develop a visual network language9) Conclusion10) Some fun exampleshellipif timehellip(ha)
Visual Network Thinking Outline
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Introduction
From the beginning network science has been a visual science rich with strange images purporting to illuminate our understanding of the world
1) Introduction2) A Little Historyhellip3) Why visualize networks at all4) What makes a good (scientific) visualization5) Network Visualization Strategies
1) Common heuristics2) Layout Features3) Static dynamic interactive
6) Guiding Idea Build theory into visualization7) Common Challenges
8) Can We Develop a visual network language9) Conclusion10) Some fun exampleshellipif timehellip(ha)
Visual Network Thinking Outline
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
1) Introduction2) A Little Historyhellip3) Why visualize networks at all4) What makes a good (scientific) visualization5) Network Visualization Strategies
1) Common heuristics2) Layout Features3) Static dynamic interactive
6) Guiding Idea Build theory into visualization7) Common Challenges
8) Can We Develop a visual network language9) Conclusion10) Some fun exampleshellipif timehellip(ha)
Visual Network Thinking Outline
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History
Eulerrsquos treatment of the ldquoSeven Bridges of Kronigsbergrdquo problem is one of the first moments of graph theory where he contrasted the observed maphellip
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History
hellip to an alternative to help think through what would became the first theorem in graph theory and the development of elements (land massesvertices) linked as pairs (edgesbridges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History
hellip and abstracted from the observed elements (land massesbridges) to abstract connections (Verticesedges)
Euler 1741
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Early kinship diagrams by Morgan (1871)
A Little network Visualization History
Diagram of consanguinity Seneca-Iroquois (lineal amp 1st collateral lines ego is male)
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Or early representations of organizational relations (1921)
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
but Morenorsquos sociograms from Who Shall Survive (1934) were something special They ignited a wave of research that was to fundamentally transform social science
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
Promising Foundations
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
With a full write-up of the 1934 medical conference in the New York Times his work sparked wide interesthellip
A Little network Visualization History
The study of networks has depended on a visual thinking since the beginning
He claims it was covered in ldquoall the major papersrdquo but I canrsquot find any other clipshellip
Promising Foundations
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Lundberg amp Steel 1938 ndash Using a ldquoSocial Atomrdquo representation
Morenorsquos idea was quickly taken up in community studies as the social analog to ldquoatomsrdquo in physicshellip
A Little network Visualization History Promising Foundations
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Charles Loomis ndash 1948
The sophistication included layering of informationhelliphere using size position and a pie-chart to tap multiple variables in addition to network structure
A Little network Visualization History Promising Foundations
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Northawayrsquos ndash ldquoTarget Sociogramsrdquo
Researchers increasingly sought ways move from artistic production to science -- The ldquotargetrdquo sociogram was an early example
A Little network Visualization History From Art to Sciencehellip
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
complete with ldquohardwarerdquo to help in the construction
A Little network Visualization History From Art to Sciencehellip
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Whatrsquos old is new Pajekrsquos fixed circular grid layout
A Little network Visualization History From Art to Sciencehellip
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Another attempt at regularization add meaning to the y and x axes (here social status and ldquogrouprdquo respectively)
Attempts at generalizing dimensional meaning start to get muddy as the figures no longer carry an intuitive connectivity pattern
Loomis amp McKinney 1956 AJS
A Little network Visualization History From Art to Sciencehellip
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Levine 1979
hellipand moves out of simple Cartesian space
The flow of images continued over time marking a wide range of potential styleshellip
A Little network Visualization History From Art to Sciencehellip
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The flow of images continued over time marking a wide range of potential styles
The space-bending challenge is made clear in modern attempts at mapping large-scale networkshellip
A Little network Visualization History From Art to Sciencehellip
Walrus (Tamara Munzer amp Caida)
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Sadly we are now starting to see caricatures of ldquonetworkrdquo images everywhere
A Little network Visualization History From Science to Sillyhellip
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A New York Times graphic on links between wall-street executives and the business school they attendedhellip
A Little network Visualization History From Science to Sillyhellip
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History From Science to Sillyhellip
hellipOr conspiracy models for disgraced politicianshellip
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History From Science to Sillyhellip
From silly to dangerous the ldquocausal modelrdquo network for the Afghan conflict
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
A Little network Visualization History From Science to Sillyhellip
hellipor downright spoofs of the real thinghellip
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Lest we think only journalist do things with marginal scientific information valuehellipmany (most) published network images are information thin
Such images often provide extremely little information and the information one can derive from the image is often wrongly-weighted
Our challenge today is two-folda) What is it about visual
representations that seem so important to network studies
b) How do we do it well
Science Art or Craft
A Little network Visualization History From Science to Sillyhellip
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The historic reliance on imagery is clear in Morenorsquos work
If we ever get to the point of charting a whole city or a whole nation we would have hellip a picture of a vast solar system of intangible structures powerfully influencing conduct as gravitation does in space Such an invisible structure underlies society and has its influence in determining the conduct of society as a whole
JL Moreno New York Times April 13 1933
We literally make the invisible visible Part of the captivation of network science is the ability to see into a world that is otherwise extremely difficult to wrap our heads around
A Little network Visualization History Lessons from history
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Seeing is believinghellip
A Little network Visualization History Lessons from history
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
hellipbut our eyes can be deceived
A Little network Visualization History Lessons from history
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
bull Simple principles of graphic design as per Tufte
-Clarity scaling layering of information avoiding ldquochartjunkrdquo effective use of color and so forth
- Challenge here mirrors cartography how much information can one effectively layer into a network diagram
This is hard to do well and even harder to automate and thus requires an active craft hand in graphic production
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Which social space
bullExample Networks to represent social space but
bull Simple space is rarely accurate so how best to distort
bull How many dimensions are relevant
bull Do the dimensions mean anything
Communication Challenges in Network VisualizationTypes of problems
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
Network specific graphic problem which social space
First two eigenvectors 3d force-directed
Types of problemsCommunication Challenges in Network Visualization
>
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problem Multidimensional data structure requires element selection
bullFor example
Global patterns or local positionsbull Are we trying to show the distinctiveness of nodes or the
general topology of the network
Communication Challenges in Network VisualizationTypes of problems
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Lessons from history
So what distinguishes a scientifically effective visualization
bullNetwork specific graphic problem Local or Global
Reduced Block-Model Labeled Sociogram
Communication Challenges in Network Visualization
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
bullNetwork Specific graphic problemsbull Model for social spacebull Resolution (local or global)bull Multiple relationsbull Dynamicsbull Densitybull Node-level attributesbull Edge-level attributesbull Hierarchical orderingbull Diffusion Potentialbull Exploration or communication
Communication Challenges in Network VisualizationTypes of problems
VisualizationStrategy
Knowledgeproblem
Data Fidelity
There is a set of basic tradeoffs made with every visualization between data the research problem and the final image which can challenge our basic instincts as scientists since we may have to make stuff up or throw away data
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes Information amp Communication
bull Helping build intuition about the social process generating the network
bull Succinctly capturing high-dimensional properties of the network
bull Being worth the spacendash the information content of the image needs to reward deeper inspection
bull Beauty ndash powerful graphs are mesmerizing
What makes a good visualization
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The simplest answer is that an image should clearly communicate something that we would have difficulty knowing any other way
This includes
Sciencebull Ability to replicate results ndash same data should produce same picture
bull Quantification ndash most features evident from a graph should be translatable to a measurement of the graph (though this may not be initially evident)
bull Theory-relevant either gives us something new to theorize about or is useful as a theory-building device itself
This probably implies a move toward problem-specific families of visualizations
What makes a good visualization
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
N=11Mean of Y = 75Regression Equation Y = 3 + 5(X)SE of slope estimate 0118 T=424Sum of Squares (X-X) 110Regression SS 275Correlation Coeff 082
These 3 series seem very similar when viewed statistically
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
0
5
10
15
0 5 10 150
5
10
15
0 5 10 15
And given these statistics we expect a relation something like this ndash a strong correlation with ldquorandomrdquo error
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
0
5
10
15
0 5 10 15
but this would also fithellip
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
0
5
10
15
0 5 10 15
hellipor thishellip
Or many more
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Visualization allows you to see the relations among elements ldquoin the wholerdquo ndash a complete macro-vision of your data in ways that summary statistics cannot This is largely because a good summary statistic captures a single dimension while visualization allows us to layer dimensionality and relations among them
0
5
10
15
0 5 10 150
5
10
15
0 5 10 150
5
10
15
0 5 10 15
While the history is deeply rooted in visual analysis why bother If we have good metrics why not use them instead
Why Visualize Network at allInsights from scatter plots
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Permuted View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
12 6 0 112 1 13 7 4 3 15 10 14 9 58
Technically all the information is retained - one could reconstruct the data pairs from both images ndash but the re-ordered panel provides no additional information
But consider changing a key assumption of the scatter-plot the scaled ordering of the axes
Standard View
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Why Visualize Network at allInsights from scatter plots
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Original White amp Harary 2001
The exact same data presented in press in distinct ways Wersquod never see this with a scatter plot
Kolaczyk Eric D Chua David B Bartheacutelemy Marc (2009)
3 representations of the same data
Why Visualize Network at allInsights from scatter plots
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Why Visualize Network at allInsights from scatter plots
Scatter plots are good models for network imagery
bull Multiple dimensionsbull Layered informationbull Global patterns with local
detailbull Shape matters
Fails as a model in a key respect The coordinate space is rigid
The reason to visualize is that we discover elements we didnrsquot already know and can communicate it effectively Gapminder
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Benefits to network graphics bull Make visible the invisible
Give a sense of reality to unobserved features Can be truly beautiful (when well done) capturing audience imaginations Can communicate complex features quickly
bull Provide multi-dimensional insights into setting processes Metrics capture a single dimension We care about the shape of related entities Help us contextualize and make sense of the metrics we calculate
bull Help us think by organizing arranging and abstracting generals from particulars Networks are already massive abstractions ndash removing qualitative micro-
detail to capture interconnectivity but we often need to abstract further to identify elements we care about
Why Visualize Network at allProblem summary
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Challengesbull No consistent or obvious display frame
Arrangement guided by heuristic principles which may not be consistent with each other
an inability to judge the ldquofitrdquo of an image to the data and small hopes of replication
bull Too much information We have multiple levels (node edge community network) and multiple
characteristics at each level necessitates ldquodenying the datardquo in favor of insights building the theory into the image
bull Scale ldquoMagnituderdquo ndash Size of the network makes standard approaches slow and
likely not useful even when fast
ldquoTemporal Scalerdquo ndash how to bind discrete occurrences to generate a structure from particulars
Why Visualize Network at allProblem summary
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
These problems feed an entire field of graph-visualization heuristics guides to tell computers how ldquobestrdquo to arrange points which focus on
bull Even node distributionbull Consistent edge lengthsbull Avoid line crossingbull Emphasize symmetrybull Maximize edge-edge anglesbull Avoid lines passing over nodes bull Avoid node overlapsbull And many many morehellip
Auto layouts have two conflicting problems to solve (a) finding a good layout and
(b) finding it quickly
What makes a good visualizationSolutions
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
VisualizationStrategy
Knowledgeproblem
Data Fidelity
Visualization as craft My instinct is that hard-and-fast rules will be difficult to find craft production is thus a better model than either art (too expressive) or technical drawing (too rigid)
Letrsquos work through some exemplars and see how they balanced these trades
Unlike a scatter-plot the default layout styles of most off-the-shelf programs will lead to very different layouts
What makes a good visualization
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Tree-Based layouts
Most effective for very sparse regular graphs Very useful when relations are strongly directed such as organization charts cluster tress or internet connections
ldquoForcerdquo layouts
Most effective with graphs that have a strong community structure (clustering etc) Provides a very clear correspondence between social distance and plotted distance
Two images of the same network
(good) (Fair - poor)
Basic Layout Heuristics
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Tree-Based layouts Spring embedder layouts
Two layouts of the same network
(poor)(good)
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
ldquogoodrdquo = high correlation between spatial distance and network distance
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit 0528 Fit 682
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Fixed Coordinate Layouts
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=057Fit=026
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Fixed Coordinate Layouts
Two layouts of the same network
Basic Layout Heuristics
Chord diagrams are the network equivalent of pie charts
Fixed Coordinate Layouts
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Use maps judiciouslyBasic Layout Heuristics
Does this network map of Facebook tell us anything more than population density
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Use maps judiciouslyBasic Layout Heuristics
Hear goal was to show that network distance spanned vast geographic distance
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Levels of distortion
Two layouts of the same network
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Shrink each cluster to accent separationFit=075Fit=081
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Network visualization helps build intuition but you have to keep the drawing heuristic Here we show the same graphs with two different techniques
Basic Layout Heuristics
Fit=081
Center fisheye on dense cluster
Fit=074
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use curves sparinglyhellipthey are usually useless ndash just extra ink ndash itrsquos a hallmark of the default Gephi stylebut doesnrsquot add much
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
3d is rarely usefulhellipalmost never in 2d presentations sometimes interactively
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Secondary style elementsSmall multiples
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
C45 B 85
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
C83 B 36
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Based on work supported by R21-HD072810 (NICHD Moody PI) R01 HD068523-01 (NICHD Moody PI) R01 DA012831-05 (NIDA Morris Martina PI)
C99 B 86
Network Diffusion amp Peer InfluenceA closer look at emerging connectivity
>
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Secondary style elements
Some seemingly innocuous features can shape the effect of network diagrams
Use color amp size to reinforce positional elementsNode shapes are almost impossible to discern
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Static Dynamic amp Interactive
Lots of new tools for interactive network visualization Consider
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Static Dynamic amp Interactive
js Searchable
httpwwwsocdukeedu~jmoody77ChefNetsnetwork
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Consider as a good example Ken Frankrsquos drawing program
The ldquorawrdquo network is distorted by emphasizing within-group ties This is then highlighted by edge-patterns (solid or dotted) and group boundaries
Guiding IdeasBuild theory into the visualization
Group-weighted MDS
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The scientists first rule is to never interfere with data ndash but this is probably exactly what is needed for (social) network science
Or the dynamic visualizations of McFarlandrsquos school classrooms
Source Moody James Daniel A McFarland and Skye Bender-DeMoll (2005) Dynamic Network Visualization Methods for Meaning with Longitudinal Network Moviesrdquo American Journal of Sociology 1101206-1241
Black ties Teaching relevant communicationBlue ties Positive social communicationsRed ties Negative social communication
Guiding IdeasBuild theory into the visualization
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Source Bender-deMoll amp McFarland ldquoThe Art and Science of Dynamic Network Visualizationrdquo JoSS 2006
The same data produce different impressions if you change the observation window
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
Guiding IdeasBuild theory into the visualization
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The scientists second rule has to be to look for regularity and exploit that for theory Consider as a good example Harrison Whitersquos Kinship model
A powerful reduction of complexity but at the cost of a particular perspective Mom = Fatherrsquos Wife for example
Guiding IdeasBuild theory into the visualization
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Abstraction helps ndash consider larger networks
Basic Problem 1 DensityScale
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
2009 Social Science Journal co-citation similarity
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Borrett Stuart R James Moody amp Achim Edelmann 2014 ldquoThe Rise of Network Ecology Maps of the topic diversity and scientific collaborationrdquo Ecological Modeling (DOI 101016jecolmodel201402019)
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
The traditional network diffusion image is the logistic curve Consider Coleman Katz and Menzel (1957)
These curves count the cumulative number of ldquoadoptersrdquo by time but donrsquot express much about network structure because they aggregate across all transmissions
That is they show an aggregate outcome without really identifying a social process
Basic Problem 2 Spread over a dynamic network
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
An alternative is to think of a network trace as the likely number reached from any given node where ldquolikelyrdquo is determined either by an edge transmission p(transfer|dist) or simple distance (geodesic) and follow each starting node
Trace Curves Number reachable at each step
The collection of curves is useful to either compare to a null hypothesis or to contextualize an observed epidemic Curves are simple to use as we have good statistical measures of their shape
But they are still one-step removed from the network itself so how might we incorporate this information into the network
Basic Problem 2 Spread over a dynamic network
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
But is hard when the graph is not simple
How might we capture the relevant aspects of diffusion in a network of gt10000 nodes
Non-adopter
Source
If your goal is to map an observed transmission process across a fixed network node coloring can be effective
Basic Problem 2 Spread over a dynamic network
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Start with Context Next a snapshot of a single day (limited to time-relevant)
Basic Problem 2 Spread over a dynamic network
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Space-Based layout of the transmission flow shows diffusion penetrates the full space but little elsehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Tree Based Layout
Since concurrency is an edge property we color the edges by the concurrency status of the tiehellip
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Tree Based Layout
but since the path carries the infection we highlight the cumulative effect of concurrency by noting any place where a transmission would have been impossible were it not for an earlier concurrent edge
This is one way to deal with the importance of network dynamics
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
An alternative version here including non-transmission edges amongst the transmission set
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
An alternative version here including non-transmission edges amongst the transmission set
hellippush it Combine hierarchy and space
Start with Context Now follow only the diffusion paths
Basic Problem 2 Spread over a dynamic network
>
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Dynamics complicate the standard point-and-line representation since you cannot really follow all ldquopossiblerdquo paths
Basic Problem 3 Display Dynamic Networks
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Temporal projections of ldquonet-timerdquo
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Basic Problem 3 Display Dynamic Networks
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
Example from Skye Bender-deMoll
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Time-Space graph representationsldquoStackrdquo a dynamic network in time compiling all ldquonode-timerdquo and ldquoedge-timerdquo events (similar to an event-history compilation of individual level data)
Consider an example
a) Repeat contemporary ties at each time observation linked by relational edges as they happen
b) Between time slices link nodes to later selves ldquoidentityrdquo edges
Temporal projections of ldquonet-timerdquo
Basic Problem 3 Display Dynamic Networks
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Example that combines many Political Polarization
Consider again our senate voting network
Network is simultaneously dense (since every pair has some score) and dynamic as the network evolves over time
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Example that combines many Political Polarization
>
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
One problem is that many nodes are structurally equivalent generating ldquoduplicaterdquo information
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Color by structural equivalencehellip
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
Adjust position to collapse SE positions
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
Network is dense since every cell has some score and dynamic the pattern changes over time
And then adjust color line width etc for clarity
While wersquove gone some distance with identifying relevant information from the mass how do we account for time
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
The extent of political polarization has been increasing recently
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
How to visualize this dynamically Try a movie
Party polarization in Congress a social networks approachA Waugh L Pei J H Fowler P J Mucha and M A Porter
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Movie is swamped by global density So letrsquos combine a time-space projection with the block model
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Consider again our senate voting network
Pol
ariz
atio
n
Time
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Political Polarization
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions Developing a visual language of networks
The simplicity of Morenorsquos sociograms was balanced against a determined consistency in presentation style that allowed one to compare across settings
The aim of solid scientific visualization should be to (re-) develop such conventions
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions Developing a visual language of networks
ldquoNounsrdquo are the easy -- identifying symbols to represent known nodeedge types should be easy to do
Shape color size-for-magnitude and usually be made pretty intuitive
The tradeoff is clutter use shapes in particularly very sparingly
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions Developing a visual language of networks
ldquoStylerdquo is always evident we can (almost) as easily recognize the signature of a particular approach as one can the voice of an author
This is both a characteristic of authors and the programs the use but is typically done without reflection
A review of elements suggests some simple features to examine for style
For my eye for example most graphs use lines that are too thick and colors that are too harsh
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
So what distinguishes a scientifically effective visualization
bullBasic quantitative visualization craft (ldquostylerdquo or ldquorhetoricrdquo)
AfterBefore
Communication Challenges in Network VisualizationTypes of problems
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions Developing a visual language of networks
The real challenge is to identify the active grammar underlying a language of networks Wersquore looking for implied rules that link visual representations to social characteristics For example proximity-as-social-distance
The key value of a grammar is that dictionaries (or ldquopattern booksrdquo) collapse under the weight of observed diversity while grammars admit many new utterances
The current body of graph-heuristics are a place to start for this grammar -- they provide aesthetic ldquoparts of speechrdquo ndash but not style
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
Wedding Program
Neuro Science meta-analysis
Animated gif of dynamics of conflict in Mexico
RDS Recruitment wave
Team project interactions (managers in green)
Slide 1
Slide 2
Slide 3
Slide 4
Slide 5
Slide 6
Slide 7
Slide 8
Slide 9
Slide 10
Slide 11
Slide 12
Slide 14
Slide 15
Slide 16
Slide 17
Slide 18
Slide 19
Slide 20
Slide 21
Slide 22
Slide 23
Slide 24
Slide 25
Slide 26
Slide 27
Slide 28
Slide 29
Slide 30
Slide 31
Slide 32
Slide 33
Slide 34
Slide 35
Slide 36
Slide 37
Slide 38
Slide 39
Slide 40
Slide 41
Slide 42
Slide 43
Slide 44
Slide 45
Slide 46
Slide 47
Slide 48
Slide 49
Slide 50
Slide 51
Slide 52
Slide 53
Slide 54
Slide 55
Slide 56
Slide 57
Slide 58
Slide 59
Slide 60
Slide 61
Slide 62
Slide 63
Slide 64
Slide 65
Slide 66
Slide 67
Slide 68
Slide 69
Slide 70
Slide 71
Slide 72
Slide 73
Slide 74
Slide 75
Slide 76
Slide 77
Slide 78
Slide 79
Slide 80
Slide 81
Slide 82
Slide 83
Slide 84
Slide 85
Slide 86
Slide 87
Slide 88
Slide 89
Slide 90
Slide 91
Slide 92
Slide 93
Slide 94
Slide 95
Slide 96
Slide 97
Slide 98
Slide 99
Slide 100
Slide 101
Slide 102
Slide 103
Slide 104
Slide 105
Slide 106
Slide 107
Slide 108
Slide 109
Slide 110
Slide 111
Slide 112
Slide 113
Slide 114
Slide 115
Slide 116
Slide 117
Slide 118
Slide 119
Slide 120
Slide 121
Slide 122
Slide 123
Slide 124
Conclusions
We are not there yet nor are we usually ldquoat our bestrdquo ndash thoughtless application of canned package visualizations often create goopy messes that convey little substantive information
The solution is thoughtful application of a well-honed graphical style ndash the classic elements laid out by Tufte over 20 years ago for statistical visualization ndashtailored to the theoretical challenges of networks
