PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... ·...

32
1 PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R and C Gravlee, eds. (2014), Handbook of Methods in Cultural Anthropology. Lanham: Rowman & Littlefield, pp. 631- 657. Social Network Analysis Introduction Social network analysis is the study of the patterns of interaction between actors. Actors are typically people, such as residents of a village (Hopkins 2011), school children (Moutappa et al. 2004) or customers of a cell-phone company (Onnela et al. 2007), but they can be other things as well. Actors can be collections of people, such as companies (Connelly et al 2011), countries (Kim and Shin 2002), or primates (Kasper and Voelkl 2009). Social network analysis can be a good way to operationalize social structure on many different levels. In this chapter we review how social networks have been used, how the data are collected, and what you can do with them. For a more complete review of social network analysis there are several books and resources available. A good introductory text is Scott’s Network Analysis: A Handbook (2000). Valente’s (2010) text Social Networks and Health: Models, Methods and Applications is an accessible resource for any discipline. An online tutorial by Hanneman and Riddle (2005) uses the program UCInet (Borgatti, Everett and Freeman 2002), the most widely used social network analysis program, to introduce concepts and measures. In addition, a recent publication by Borgatti et al. (2013) provides a good introduction to the analysis of network data with UCinet, and other related programs. You will find a more in-depth treatment of social network measures in Wasserman and Faust (1994). Kadushin’s recent book Understanding Social Networks: Theories, Concepts and Findings provides a comprehensive view of the way social network analysis has been applied. Jackson’s (2008) book Social and Economic Networks applies social network concepts to economic theory and methods, such as game theory. Historical and Theory Today most people think of “social networks” as social networking sites or services such as Facebook and Twitter. The idea of social networking sites is indeed based on social network research conducted by Stanley Milgram (1967) about the “Small World problem”. Milgram and his colleagues showed that completed chains connecting two randomly selected people in the U.S. contained, on average, about five-plus people, or six degrees of separation. The first commercial social networking site, sixdegrees.com, was launched in 1997 and although it was closed a few years later it started an enormous commercial industry that has changed the way we communicate.

Transcript of PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... ·...

Page 1: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

1

PREPRINT

McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R and C Gravlee, eds.

(2014), Handbook of Methods in Cultural Anthropology. Lanham: Rowman & Littlefield, pp. 631-

657.

Social Network Analysis

Introduction

Social network analysis is the study of the patterns of interaction between actors. Actors

are typically people, such as residents of a village (Hopkins 2011), school children (Moutappa et

al. 2004) or customers of a cell-phone company (Onnela et al. 2007), but they can be other

things as well. Actors can be collections of people, such as companies (Connelly et al 2011),

countries (Kim and Shin 2002), or primates (Kasper and Voelkl 2009). Social network analysis

can be a good way to operationalize social structure on many different levels.

In this chapter we review how social networks have been used, how the data are

collected, and what you can do with them. For a more complete review of social network

analysis there are several books and resources available. A good introductory text is Scott’s

Network Analysis: A Handbook (2000). Valente’s (2010) text Social Networks and Health:

Models, Methods and Applications is an accessible resource for any discipline. An online

tutorial by Hanneman and Riddle (2005) uses the program UCInet (Borgatti, Everett and

Freeman 2002), the most widely used social network analysis program, to introduce concepts

and measures. In addition, a recent publication by Borgatti et al. (2013) provides a good

introduction to the analysis of network data with UCinet, and other related programs. You will

find a more in-depth treatment of social network measures in Wasserman and Faust (1994).

Kadushin’s recent book Understanding Social Networks: Theories, Concepts and Findings

provides a comprehensive view of the way social network analysis has been applied. Jackson’s

(2008) book Social and Economic Networks applies social network concepts to economic theory

and methods, such as game theory.

Historical and Theory

Today most people think of “social networks” as social networking sites or services such

as Facebook and Twitter. The idea of social networking sites is indeed based on social network

research conducted by Stanley Milgram (1967) about the “Small World problem”. Milgram and

his colleagues showed that completed chains connecting two randomly selected people in the

U.S. contained, on average, about five-plus people, or six degrees of separation. The first

commercial social networking site, sixdegrees.com, was launched in 1997 and although it was

closed a few years later it started an enormous commercial industry that has changed the way

we communicate.

Page 2: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

2

Apart from this commercial meaning, “social networks” and “social network analysis” has

a long history. In his account of the development of the field Freeman (2004) shows how the

roots of social network analysis can be traced simultaneously to the work of Jacob Moreno

(1934) and anthropologists from the Manchester School. Moreno represented social interactions

as dots for actors and lines for ties between actors, a procedure he called Sociometry. His work

influenced British anthropologists of the 1940s and ‘50s. The structural-functionalist paradigm

had served until then for the study of single, spatially compact, small-scale, but something else

was needed to understand trying to understand new patterns of social interaction among tribal

people living in urban settings under colonial rule (Mitchel 1972, Wolfe 1978).

In 1947 Max Gluckman, former director of the Rhodes-Livingstone Institute in Northern

Rhodesia, started a series of seminars at the University of Manchester. Gluckman was

interested in explaining conflict and change in Africa, taking into account these new social

relationships among individuals from different cultures that emerged with British rule. During the

development of these seminars social networks ideas represented a way to bring individual

agency into the analysis, representing a source of change and intercultural variation through

competition and individual exchange strategies in complex societies.

Among the most influential scholars who attended the seminars and applied the idea of

networks to their research were J. Clyde Mitchell (social structure in Central Africa, 1969),

Elizabeth Bott (marriage and kinship in London, 1957), John Barnes (decision-making in

Norway, 1954), Bruce Kapferer (conflict in Zambian factories, 1972), Arnold L. Epstein (norms

and social situations in Zambia, 1969), Jeremy Boissevain (personal networks in Malta, 1974)

and S.F.Nadel (theory of social structure, 1957).

Barnes was the first to employ the term social network to characterize the set of relations

among institutions in a small fishing village in Norway. These relations bridged different

groupings and institutional settings on a personal level. This was similar to Gluckman’s idea of

the existence of bridging ties between traditional African cultures and British colonial culture.

Mitchell is credited with formalizing social network theory in a way that also suggested how it

should be studied. Among other network ideas, he suggested the term ego-centered networks

to describe the notion of recording data on the links surrounding a focal individual. In his view

these ego-centered networks combine to form a total network.

The Manchester tradition endured for another two decades but among other reasons,

the lack of methods (and, particularly, the lack of computers) for dealing with the large amount

of network data collected during fieldwork finally stopped its development. The network

approach to the study of social structure developed by Nadel in the Manchester seminars was

adopted by American sociologists like Harrison White and others in a series of seminal articles

(Lorrain and White, 1971; White et al. 1976, Boorman et al. 1976). From this point, the

University of Chicago and Harvard University began to develop the methods and substantive

applications of whole-network analysis. In 1971 Freeman started the journal Social Networks

and in 1977 Barry Wellman organized the International Network for Social Network Analysis

(INSNA), the organization at the center of the interdisciplinary field that social network research

is today.

Page 3: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

3

The first theoretical concern that motivated social network thinking was the problem of

how to identify and assess the effect of emergent social structures on individual behavior (and

norm development). These structures were studied at the egocentric level (i.e. the spread of

gossip, Epstein 1969) or at the whole network level (i.e. the mobilization of supporters in a work

setting, Boissevain 1972). This novel perspective allowed researchers to study actual groupings

of individuals in different roles instead of focusing solely on normative groups (kinship groups,

territorial groups, age groups, etc., see Srinivas and Béteille 1964).

The second theoretical question addressed by anthropologists was the correspondence

between cognitive and behavioral data. This task was undertaken by anthropologists influenced

by American cognitive anthropology, trained in formal methods for clustering classification

patterns and identifying intracultural individual variation (Johnson 1994). These contributions

from cognitive anthropology were accelerated by Borgatti’s (1987) development of Anthropac, a

set of data collection and analysis routines that made this type of data analysis accessible to a

much wider audience. Another contribution in this field was a series of articles about the

accuracy of social network data, Bernard and Killworth (1976). They found that cognitive

information about network relations showed systematic bias and differed from behavioral data.

The wealth of network behavioral data available in SNS as Facebook and twitter and its relation

with user cognitive reports makes today this research of special interest for current scholars

interested in social media.

The third theoretical concern came from the kinship studies, one of the core subjects in

classical anthropology. The question was whether kinship normative systems existed in

practice, and what were the empirical based models that accounted for effective kinship ties.

Per Hage studied ethnographic and historical cases of kinship. His collaboration with Frank

Harary, a mathematician interested in developing mathematical applications for the social

sciences, has proved to be influential in the development of the field of social networks (Hage

and Harary 1997). Douglas White and his students have produced a large corpus of knowledge

about the computational representation and analysis of large datasets of kinship data (Douglas

and Johansen 2005). They found that it was possible to find basic structures of relinking

intergroup marriage that maintained overall cohesion despite the appearance of fission

processes through time. The development of the computer program Pajek by Vlado Batagelj

and his colleagues made it possible to process these large data sets in order to test these

models (White, Batagelj and Mrvar 1999). Work in France from the Ecole des Hautes Etudes en

Sciences Sociales have extended this work (Houseman and White 1996, Hamberger 2011).

Newly available data on indigenous societies, mainly from Brazil, is fostering the development of

network applications to kinship (Barbosa de Almeida et al. 2010).

Another issue that guided anthropological research on networks is the question about

how macro (societal) and micro (personal) networks are related. Bernard, Killworth and McCarty

(Bernard et al. 1988, 1990) conducted a series of studies called the Reverse Small World

(RSW) using a personal network approach. These studies extended the work of Milgram,

showing that the links in the chains were in fact not random, and that people’s choice of links

was typically guided by thos links’ location and occupation. A related line of research

concerned a set of methods to estimate the size of hard-to-count populations. Using personal

Page 4: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

4

networks these researchers were able to estimate the network size of each respondent in a

population survey. This proved important in estimating the size of hard-to-count populations, like

people living with HIV, the homeless, women who had been raped, civilians killed in wars, and

so on (Bernard et al, 2010). Personal network research has been expanded by McCarty and his

colleagues with the development of EgoNet1, an open source program for collecting, analyzing

and visualizing personal network data (McCarty et al. 2007). In addition, the growing field of

transnational studies (Vertovec 2009), and the concepts of “transnational social field” and

“transnational social space,” are built on ideas from social network analysis (see Molina et al.

2012 for a review). These “fields” or “spaces” are emergent cross-border social structures that

allow migrants to overcome the hurdles of making a living in a destination country.

Finally, the question of factors that account for intracultural variation of ecological

knowledge has inspired research projects that apply social network methods (Berkes et al.

1998, Ellen et al. 2000). Examples include the study of medicinal plants (Hopkins 2011), the

conservation of traditional varieties of plants through the exchange of seeds (Calvet et al. 2012)

and the relation of knowledge to status and health in indigenous societies (Reyes-Garcia et al.

2008). These studies have shown that the structural position of an individual in a network of

knowledge and/or exchange is related to local ecological knowledge and to actual exchange.

Apart from these lines of research there have been contributions from individual

scholars, such as Larissa Adler-Lomnitz (1975) who applied social network analysis to a wide

range of ethnographic cases in Chile, Mexico, Russia and Hungary. Thomas Schweizer (1997)

contributed to the development of the theoretical concept of embeddedness based on

comparative ethnographies. Robin Dunbar, a psychologist, has done unique research in the

area of personal-network size and how it relates to physiological constraints of memory.

Dunbar suggests that the ability of people to know other people is constrained by time and

resources, as well as the organization of family, friends and acquaintances in memory, to an

average of 150 people. He points to the tendency of various human social groups to fission at

around the same size (see Gonçalves et al. 2011 for a review).

Much of the classical work in social network analysis is from sociologists. In 1961, J.S.

Coleman published and influential book about the divide between adolescent and adult worlds

in America, The adolescent society: The social life of the teenager and its impact on education.

Coleman pointed out the structure of peer interactions as one of the sources of normative

reinforcement (against the institution, in this case). Beginning in 1964, Edward Laumann

conducted a series of surveys as part of the Detroit Area Study (1973). Laumann included

questions stemming from his interest in social distance. Laumann was the first to collect

personal-network data about a small number of network members from a large and

representative sample of respondents using a standard survey approach, a method that

dominated personal-network analysis for decades. In 1973, Mark Granovetter developed the

idea of the importance of weak ties and social structure. In his research people often found out

about job opportunities through people they knew. What was unexpected was that these

opportunities typically came through acquaintances rather than people the respondents knew

1 (http://soucerforge.net/egonet)

Page 5: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

5

well (strong ties). This was because acquaintances had access to information about job

openings in groups that the close and highly connected network members did not have.

Around the same time as Granovetter’s research, Barry Wellman conducted a large

personal-network study in a suburb of Toronto using methods similar to Laumann’s. Wellman

and his colleagues discovered that despite the common wisdom that people in urban situations

lost their sense of community, communities still existed through sometimes distant network

linkages (1979). Claude Fischer in his book To Dwell Among Friends: Personal Networks in

Town and City (1982) explained how personal networks operated in all aspects of life, including

work, religion and attitude formation. Fischer was one of the first to introduce the idea of the

personal-network life cycle; that is, the ways network composition and structure change as we

age.

Charles Kadushin (1968) used the personal-network approach to construct whole

networks of invisible groups. He used the technique of snowball sampling to operationalize

Georg Simmel’s concept of social circles. Social circles are groups of people who share

characteristics in common, and therefore are influenced by each other either directly or

indirectly, although they may not be aware of that influence. Scott Feld was also interested in

social circles, but more so in how they actually formed. He introduced the notion of social foci,

social situations that result in non-random interaction between people. Feld used variations of

this theory to explain the attributes of personal networks and how those attributes are perceived

by individuals (1981).

Ronald Burt (2004) developed measures of social capital based on the presence or

absence of certain types of social ties. In an important work, examining the egocentric network

of managers in business settings, Burt developed the concept of the structural hole—that is, an

opportunity for individuals to act as bridges in otherwise disconnected groups within an

organization. Burt’s ideas about how networks stimulate creativity are also applicable outside

organizational settings. Tom Valente (2010) used variability of the characteristics of personal

networks to explain contraceptive use among women in Cameroon (1997). His research falls

under the umbrella of the diffusion of innovations, another large arena of network research

peioneered by Coleman (Coleman et al. 1966) and developed by Everett Rogers and his

students (Rogers 2003; Richards and Barnett 1993; Rice 1994, Richards and Barnett 1993).

More recently, Christakis and Fowler (2009) analyzed network data from the Framingham heart

study and suggested that conditions such as obesity are influenced by social networks.

Although this conclusion has been strongly criticized (Aral et al. 2009, VanderWeele 2011,

Rohila and Thomas 2011) the contributions coming from public health are important and

influential. Studies relating individual well-being and social networks include Cohen (1985),

Berkman and Syme (1979), and Kadushin (1982). Studies on the role of network structures on

the spread of infectious disease, especially AIDS, include Klovdahl (1985) and Latkin et al.

(2003).

The next sections are devoted to explaining both whole network and personal network

analysis. In general, whole network analysis is especially useful when indirect links are

supposed to play an important role in the expected outcome or when the behavior under study

Page 6: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

6

is expected to be at least partly explained by some institutional context (as the work place or a

retirement center, for instance). By contrast, personal network analysis is better suited for

studies in which a representative sample of the population is needed or when the whole

complex of institutional context in which ego is embedded is supposed to be important for a

given outcome (like smoking or evincing some ethnic identity). Both types of analysis can be

combined as well, as it will be explained later.

Whole Network Data Collection

Social network analysis is often classified into two types, whole networks and ego-

centered networks. Whole networks focus on the interactions of actors within a bounded space.

Boundary definition—who will be included and who will not—is an important aspect of whole

network analysis. Boundaries may be geographic (households in a village), or social (members

of an organization). Snowball sampling, where a sample of seed respondents nominate later

respondents, is often used as a way to define whole network boundaries when the network to

be studied is known (such as a gang) but where network members are not limited to a known

geography and there are no lists of network members.

With a whole network study the objective is to collect data about each member of the

bounded group and their interactions with all other members. Interactions can be measured in

many ways. A common approach is to ask members about their interactions with other

members using a scale. For example, respondents might report their closeness to every other

member on a scale of 0 to 5 where 0 implies they are not at all close and 5 implies they are as

close as possible (see Figure 1).

Figure 1. Matrix representing interaction between all pairs of members of a class.

The first cell intersecting David with David is a 5, reflecting the fact that David is as close

to himself as he can be. This is the case for all cells on the diagonal and these values are

typically ignored in the analysis of social network data. The second cell in the first column

represents Faith’s assessment of her closeness with David which she evaluated as a 1. The

first cell in the second column is David’s assessment of his relationship with Faith, which he

evaluated as a 2. Faith and David do not agree on their relationship. The cells in the bottom

half of the matrix (those numbers below the diagonal) are different from those in the upper half.

This means that the matrix is asymmetric. If the upper and lower half were the same the matrix

David Faith Rosanna Antonio Napp Lem Jim Beth Mark Kent Amber Thomas

David 5 2 2 0 0 1 0 3 1 0 2 0

Faith 1 5 5 0 0 0 0 1 0 0 2 0

Rosanna 2 5 5 0 0 1 0 2 0 0 4 0

Antonio 0 1 1 5 0 0 0 0 0 0 0 0

Napp 0 0 0 0 5 0 0 0 0 0 0 0

Lem 2 0 2 0 0 5 5 2 0 0 2 0

Jim 0 0 1 0 0 5 5 5 0 0 2 0

Beth 4 3 1 0 0 1 5 5 0 0 3 0

Mark 0 0 0 1 0 0 0 0 5 0 1 0

Kent 0 0 0 0 0 0 0 0 0 5 0 3

Amber 2 3 3 0 0 1 2 2 1 0 5 0

Thomas 0 0 0 0 0 0 0 0 0 3 0 5

Page 7: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

7

would be symmetric. Symmetry is an important concept in the analysis of network data and

must be considered in designing a data collection and analysis strategy. Many network ties are

not reciprocal, such as closeness (in Figure 1). Exchange of resources can be asymmetrical (A

may give more to B than B does to A) or symmetrical (if the question is “Do A and B exchange

resources at all?”)

In the example above only one question represents the tie between actors. Social

networks may be represented by many tie definitions, each one potentially resulting in a

different network structure. Table 1 provides a list of commonly used questions for tie definition

in anthropological studies. Each definition is not one question, but one for every member of the

group a respondent is evaluating, creating the potential for excessive respondent burden.

Anthropologists should be aware of the potential for low variability of responses to some tie

questions, such as whether members of a small village know each other. Although the last two

questions in Table 1 (referring to duration and frequency) may be appropriate in some

situations, they are not necessarily correlated. For example, the people someone interacts with

frequently may not be those they have known the longest. Similarly, the people someone sees

most often may not be the closest to them emotionally. Tie questions must be designed

carefully to reflect the network interaction that aligns with the theory and specific circumstances

of an individual study.

How close are you to X?

How likely are you to borrow money from X?

How much do you rely on X for information?

How much labor do you exchange with X?

How much meat do you exchange with X?

Do you exchange tools with X?

How long have you known X?

How often to you interact with X?

Table 1. Example tie questions.

Ideally one would collect responses from every member of the whole network (all

villagers or all members of an organization) since respondents may not agree about their

patterns of interaction. However, as networks become larger it becomes difficult to get

responses from all network members. The need to measure a network within a fixed period of

time may not be possible without assistance. For example, it may be possible to collect

interaction data within a village across several years, but in that time the structure of interactions

can change—in which case, the tie data collected earlier do not reflect the structure when the

data are analyzed.

One of the most widely used approaches for collecting whole network data is the roster

method where every respondent evaluates their interactions with every other member on a list.

If the list is relatively small, say under 100 members, this is the most accurate and

recommended approach. Showing each member of a group the names of all other members of

the group triggers information retrieval about relationships. Another approach is called the

survey method where respondents are asked to list the members they interact with.

Respondents might be asked to list five households they rely on for water during a drought.

Page 8: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

8

This approach has the disadvantage that respondents may forget subtle connections within the

whole network, but it is much faster and less burdensome than the roster method. This method

is recommended when the network is large. Some researchers do not limit the number of

nominations when using the survey method, instead allowing respondents to choose the

number of network members. While it is almost certain that the number of households one can

trust with child care varies from one household to another, it is not certain that the number

nominated actually reflects that. Some respondents may list few members because they are

uncooperative or simply forget. The researcher has to balance these two possibilities. When

the boundary definition results in a very large network it may not constitute a meaningful one.

For example, the city of Mumbai, India may not comprise a single network in any meaningful

way.

There is some tolerance for missing data in whole network analysis, particularly when

the data are asymmetric. While a single respondent may refuse to be interviewed, other

respondents will report on their interactions with the missing respondent. Research suggests

that for some structural measures as little as 50% of responses is sufficient to capture network

structure (Costenbader and Valente 2003).

As mentioned earlier, the network being studied need not be of people; it can be a

collection of people. Anthropologists conducting a community study may want the network to

reflect the interactions among community members, particularly when age and gender roles are

a consideration. For other studies it may be more appropriate to use households as the unit of

interaction, such as when studying networks of exchange. Even when the household is the unit

of analysis the interviewer will have to talk to someone. Choosing who to talk to about the

interactions with the household may have consequences as a single person may not know

about all interactions, and those they know about may be biased. One approach is to randomly

select respondents from the household, an approach often used in survey research. This will

have the effect of eliminating systematic bias, but it is possible the randomly selected

respondent will not know about the actual interaction of all members. Another approach is to

select the person responding for the household who is believed to be the most knowledgeable

about certain types of interaction. Still another approach is to try to achieve a consensus

opinion among a group of household members characterizing the interactions with other

households. This approach may be important when there are highly varying types and amounts

of interaction among household members that a single member could not report accurately.

Sometimes network data can be collected by observing interactions rather than asking

people to report on their interactions. Studies of interaction among preschoolers rely on trained

observers who carefully record their interactions (Martin et al. 2005). This approach works

when there are a relatively small number of actors to observe, or when there are many

observers. When using an observation approach, coordination of those observations is

important to maintain efficiency. Another approach is to sample times of day and specific

locations to observe interactions. Over time, recurring interactions should be observed more

frequently.

Page 9: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

9

Some data on interaction may exist in an archive. For example, in office environments

e-mail and phone records may be used as measures of interaction. In villages, marriage

records may indicate interaction between households. Historians have used records of financial

transactions to measure networks of exchange over time (Padgett and Ansell 1993). One could

measure the width of roads and paths between households as a way to operationalize

interaction. Existing data should always be considered as an alternative or complement to data

that requires direct interviewing of respondents.

The data depicted in Figure 1 are referred to as one-mode data, meaning the labels

(mode) in the rows are the same as the labels in the columns. Another convenient way to use

archival data is with two-mode data collection. With two-mode data the labels in the rows are

not the same as the labels in the columns. Typically the rows represent the actors and the

columns represent some event linking the actors. Figure 2 shows a hypothetical two-mode

network of a set of villagers and the funerals they attend. Row 1 indicates that Buseje attended

only funerals 5 and 6 while Alile attended funerals 1 and 2. Assuming that co-attendance

implies some form of social interaction, if two people attend the same funerals they may

interact.

Figure 2. Two-mode data of funeral attendance.

The data in Figure 2 can be transformed into a one-mode network as seen in Figure 3.

This one-mode network is symmetric, meaning the upper and lower halves are the same. The

numbers down the diagonal represent the number of funerals the person attended; Buseje

attended two funerals and Chisulo attended three. The numbers in the off-diagonal indicate the

number of funerals the actors attended in common. Buseje and Chisulo did not attend any of

the same funerals while Buseje and Dziko attended two of the same funerals. A two-mode

network approach is also useful for collecting data directly from respondents. For example,

rather than observing funerals an interviewer could ask respondents to indicate if they had

attended each of a set of funerals.

F1 F2 F3 F4 F5 F6

Buseje 0 0 0 0 1 1

Chisulo 1 1 1 0 0 0

Mandala 0 0 1 0 0 0

Kapeni 0 0 0 1 0 0

Dziko 1 1 1 1 1 1

Mwai 0 0 0 0 0 1

Alile 1 1 0 0 0 0

Page 10: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

10

Figure 3. One-mode affiliation from two-mode data.

The data collection strategy will also depend on data entry constraints from analysis

software. Network data are typically analyzed as a matrix, such as those depicted in Figures 1

and 3. Although it is possible to type data into a matrix format if the number of actors is small, it

will be time-consuming and error-prone for larger networks. It is also not necessary. Most

social network programs, such as Ucinet and Pajek

1, accept data collected in different ways. Two of the most frequently used methods are an

edge list and a node list.

Social network analysis borrowed many concepts from graph theory. In the jargon of

graph theory, the actors are nodes and the ties between actors are the edges. As an alternative

to filling in all the cells in a matrix, one can list all the ties (edges) that are recorded or observed.

Figure 4a shows an edge list and Figure 4b shows the resulting matrix.

Figure 4a. Example of edge list.

Figure 4b. Matrix produced from edge list.

A node list is the best approach when entering data collected using the survey method

where respondents list actors to whom they are tied. Figure 5a shows what a node list might

Buseje Chisulo Mandala Kapeni Dziko Mwai Alile

Buseje 2 0 0 0 2 1 0

Chisulo 0 3 1 0 3 0 2

Mandala 0 1 1 0 1 0 0

Kapeni 0 0 0 1 1 0 0

Dziko 2 3 1 1 6 1 2

Mwai 1 0 0 0 1 1 0

Alile 0 2 0 0 2 0 2

Juan Anna 2

Juan Melecio 4

Anna Melecio 1

Paulo Juan 3

Paulo Melecio 2

Maria Anna 5

Maria Melecio 2

Juan Anna Melecio Paulo Maria

Juan 0 2 4 0 0

Anna 0 0 1 0 0

Melecio 0 0 0 0 0

Paulo 3 0 2 0 0

Maria 0 5 2 0 0

Page 11: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

11

look like. The first name in each row is the respondent and the names following are the ones

they list.

Figure 5a. Example of node list.

Figure 5b. Matrix produced from node list.

Network visualization

One of the first things researchers do with social network data is to create a network

visualization. The data from Figure 1 are visualized in Figure 6 using a program called

Netdraw2. Visualization gives a sense of the overall network structure. In Figure 6 the dots

(nodes) represent the students and the lines indicate how they evaluated each other in terms of

closeness. The thickness of the line indicates the level of closeness and the arrow (present or

absent on each end of the line) indicates if the tie was reciprocated.

In order for these data to be visualized the software must decide whether a line will be

drawn. The data from Figure 1 are asymmetric, meaning that the students did not necessarily

agree about how they were tied to one another. The default for most visualization programs is

to draw a line that represents the maximum evaluation. For example, David rated Faith a 2

while she rated him a 1. The program must show a line or not, so it defaults to the maximum of

2. David rated Mark a 1 but Mark rated David a 0. The program inserted a line. Typically

network analysts will use social network software to symmetrize the data to control how those

choices are made. Many social network measures are graph-based and require dichotomized

(binary) data. If the data are not already dichotomous the software will typically default to any

value greater than zero.

Juan Anna Melecio

Anna Melecio

Paulo Juan Melecio

Maria Anna Melecio

Juan Anna Melecio Paulo Maria

Juan 0 1 1 0 0

Anna 0 0 1 0 0

Melecio 0 0 0 0 0

Paulo 1 0 1 0 0

Maria 0 1 1 0 0

Page 12: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

12

Figure 6. Visualization of class network.

There are many ways to visualize the connections between actors. In Figure 6 the data

are visualized in two dimensions using a program called Netdraw (Borgatti 2002). Other

programs such as Mage2 can create three-dimensional visualizations which may be more

appropriate for some types of data. For example, if there were three primary drivers of social

interaction, such as sex, age and income, the two-dimensional visualization would only show

two. Most social network visualizations are in two dimensions, simply because two dimensions

are easier to visualize and to publish.

One can choose from several techniques to position the actors within the visualization

space. Techniques such as circle graphs do not focus on the position of the node at all, but

rather the clustering of lines (Freeman 1999). Other techniques are far more complicated.

Another consideration is whether the visualization is based on valued data or dichotomized

2 http://kinemage.biochem.duke.edu

Page 13: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

13

data. Statistical procedures such as multi-dimensional scaling (MDS) position the nodes using

coordinates that incorporate the value of the tie between the nodes.

Figure 6 uses a procedure called a spring embedder that relies on the presence or

absence of a tie (dichotomous data). A spring embedder selects a node at random and places it

in the middle of the screen. It selects a second node at random and if it is tied to the first node it

puts them next to each other, and if they are not tied it pushes them apart. This procedure

continues until all nodes are optimally pulled towards those to which they are connected and

pushed apart from those they are not. Some programs, like Netdraw, take nodes that are not

tied to any other node (isolates), and stacks them in one place on the graph, such as the upper

left. Other programs randomly place isolates in the two-dimensional space. All programs

randomly place disconnected groups (i.e. components) in the two dimensional space.

When interpreting network visualizations the length of a line is typically not important. If

a line exists between two nodes it means they are tied. A longer line means that the two nodes

(the dyad) tend to be less tied to the same people than two nodes connected by a shorter line.

One must be careful about interpreting the position of one component relative to another

component. If two groups do not have any ties then their placement in the two-dimensional

space is random.

Network visualizations often incorporate the attributes of the actors. For example, one

can use color, labels, size or shape to show how the network structure represented by the

visualization can be explained. Some attribute data will come from data collected about each

actor. Other attributes, such as network measures, can be calculated using software.

Many researchers use network visualizations as a cue to talk to respondents about their

group. Network visualizations tend to be intuitive to respondents. Network structures can

suggest questions such as why some actors are isolated from others, why some people tend to

be very central and why some groups are separated. These are questions that would be

difficult to frame or to explain to respondents without the aid of network visualization (Molina et

al. forthcoming).

Social Network Measures

Although visualization is a great way to get a feel for your data, there is no substitute for

objective measures. Social network measures can be broken into four types as depicted in the

quadrants shown in Table 6.

Graph-Based Statistical

Group Level Centralization Number of Components Number of Isolates Density

Quadratic Assignment Procedure

Node Level Centrality Cluster Analysis Multi-Dimensional Scaling

Table 6. Types of social network measures

Page 14: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

14

The rows refer to whether the network measure is a single number representing the

group or an individual measure for each node within the network. Network density, one of the

most widely used group level measures, is the percent of all ties that exist out of all those that

could exist. It is represented as a proportion between 0 and 1.

Node level measures say something about the position of the actor within the entire

network. The most widely used node level measure is centrality. While there are many

centrality measures, the most commonly used are degree and betweenness centrality. Degree

centrality for a node is the number of other nodes it is tied to directly. In the class network

depicted in Figure 6 the degree centrality for Antonio is 3 because he is tied to 3 people.

Betweenness centrality is calculated by determining the shortest path (the geodesic) between

every pair of nodes. For a given node the betweenness is the number of shortest paths they lie

on. A node that is on a lot of shortest paths may be in a brokering position. In Figure 6 Amber is

the most between central node. Centrality measures are often exported from network software

and used as covariates in statistical models. For example, betweennness centrality could be

used to explain the variability in a measure of social mobility.

Actors with high centrality do not necessarily use that quality to their advantage. Actors

can be central for many reasons. They may be important, or they may be central because they

are exploited by others. They could even become central if they were mutually disliked by most

people in the network. When interpreting the structure of a group it is valuable to consider why

some people are marginalized, or others are embedded in closely knit groups.

Many social network measures are based on dichotomous data. However, there are

many statistical measures that can be used for analyzing valued matrices. For example, cluster

analysis uses valued matrices to aggregate nodes into subgroups. Finding subgroups is often

an objective of social network analysts, and statistical procedures are useful for doing that.

Here is a list of commonly used social network measures with a brief description:

Degree Centrality – An alter is highly degree-central to the extent he or she is directly connected

to many other alters.

Degree Centralization – A measure of the extent to which the network is dominated by a few

degree central nodes.

Closeness Centrality –An alter is highly close-central if he or she is connected by short paths to

most other alters.

Closeness Centralization – A measure of the extent to which the network is dominated by a few

close central nodes.

Betweenness Centrality –A node is highly between-central to the extent he or she lies on many

geodesics (shortest paths) between alters. Alters that are between central are brokers.

Betweenness Centralization – A measure of the extent to which the network is dominated by a

few between central nodes.

Page 15: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

15

Components – A set of alters who are connected to one another directly or indirectly.

Isolates – A node unconnected to any other node.

Another category of measures focuses on the tie (edge) rather than the node. These are

important when the research question has more to do with the relationship than with individual

actors. For example, a study of exchange networks within a community could focus more on

the structural properties of a tie between two actors, such as edge betweenness, rather than the

structural position of either actor.

Social Network Models

The measures described in the previous section result in single numbers to represent

the network or the nodes in the network. These measures, particularly the node level

measures, can be used in statistical models to predict, or be predicted by, other variables. A

common strategy is to calculate the degree or betweenness centrality of each node and to use

that in a model to explain the variability of a dependent variable.

In contrast to a statistical model is a network model. Network models determine the

fundamental relationships that underlie network structures. In social network analysis there are

models that rely on the distribution of different types of triads within the network. A triad is a set

of three nodes, and these three nodes can exhibit any of 16 different types of relationships

depending on whether the nodes are tied, and if they are tied whether the relationships are

reciprocated. An examination of the triads in a network is called a triad census. Some network

analysts consider triads to be the building blocks of social structure (Wasserman and Faust,

1994; Killworth and Bernard 1979; Johnsen 1985).

A level up from triad analyses are blockmodels. Blocks are sets of interacting nodes that

are mostly tied relative to the potential ties between blocks. Social network software can be

used to identify the optimal number of blocks, or subsets, within a network (see de Nooy et al.

2005 for the software Pajek). Another set of models are p1, p2 and exponential random graph

models (ERGM, see Wasserman and Pattison 1996; Wasserman and Robins 2005; Robins et

al. 2005). These are used to understand what explains network structure – the attributes of the

nodes or the structural position of the nodes. Network models are particularly useful in

analyzing longitudinal networks —that is, networks over time. Network software such as

RSiena3 and Statnet (Snijders 2005, Snijders et al. 2010) are useful for determining whether

changes in the network structure over time are due to attributes of nodes or the probability of

structural change due to their starting structural position. Longitudinal social networks, though

difficult to collect, are increasingly used to answer questions about what makes networks

change and what those changes predict. Many modeling algorithms are available for free in the

R development environment.

Ego-centered Networks

Page 16: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

16

Ego-centered networks, also called personal networks, examine the social context

surrounding a focal respondent (egocentric networks refer to the analysis for every vertex or

node of the adjacent nodes’ structure and composition within a whole network). In contrast to

whole networks, ego-centered networks are not typically bounded geographically or socially.

The purpose of an ego-centered study is to operationalize individual social context as a set of

variables that are used to predict or say something about the focal respondent.

Respondents to an ego-centered network study are selected much the same way as

respondents would be selected for a survey. The selection of respondents, called egos, is

usually dictated by the group to be studied. The group might consist of people who live in a

particular place, such as a village, or certain types of people, such as the elderly. The objective

is to say something about the group from which the respondents have been selected, and the

rules governing survey sampling, such as targeting margins of error, will apply. Programs such

as Egonet4 are designed to collect and analyze ego-centered network data.

There are generally four parts to an ego-centered network study. First are the questions

one asks the respondent (ego) about themselves. These are typically the dependent variables,

such as the number of children a respondent has or whether they are a smoker, and/or the

control variables, such as age and ethnicity. Some network studies identify spurious

relationships between social network variables and the dependent variables. That is, they may

be related through a third variable which is often easier to measure. Without control variables

one cannot identify the unique explanatory power of the network variables.

The second step is to elicit the names of network members, called alters, from the

respondent. Unlike whole network studies where the network members are typically known,

with an ego-centered study the respondents list their network members. The questions used to

get those names are called name generators. Name generators provide the boundary definition

for the ego-centered network. There are many things to consider in designing name generators.

The first is that the respondent is the only one who knows who is actually in their network, and

unless alters are contacted (called alter-chasing) only respondents know if the list is correct.

This introduces the possibility of error in the elicitation from one respondent to the next. The

name generator should be worded carefully so this variability is minimized across respondents.

Table 7 shows some name generators that have been used in the past. These name

generators could result in very different lists of names. The number of people you have ever

known will be a much larger list than those you currently talk to about important matters. Both

will yield valid lists of names depending on the focus of the research.

People you talk to about important matters

People you can rely on for help with a personal problem

People who you know by sight or by name that you have had contact with in the past two years

People you have ever known

Table 7. Some name generators.

Another consideration is the bias associated with the way respondents recall names.

People do not store names in memory in exactly the same way. This introduces the potential

for bias in the way names are recalled. When given a very general name generator most

Page 17: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

17

people will start listing their immediate family, although that will not be true of all respondents.

Respondents may continue with extended family, or work colleagues, those they have seen

recently, or even names that are similar to names that have already been listed. Elicitation bias

is unavoidable. It is therefore advisable to word the name generator so that all respondents

have the same bias. For example, one could ask respondents to list people beginning with

those they have seen most recently. While this creates a biased network, at least all

respondents have the same bias.

Another design question is the number of alters to elicit. This will in part be determined

by the research topic, but also must be balanced by the potential respondent burden. Some

researchers fix the number of alters at, say 20 or 30, while others allow respondents to have

variable numbers of alters, using the number they list as a proxy for network size. There are

other methods to measure network size, such as the network scale-up method (see Bernard et

al 2010). There are many reasons someone may stop listing alters – they may have listed all

the alters, they may be fatigued, they may not remember more alters, or they may be

uncooperative. Since it is not known why a respondent stops listing alters it may be better to fix

the number of alters for all respondents, requiring that they all expend the same effort.

The next stage after alter elicitation are the questions about each alter; these are called

name interpreters. The data from these questions will result in compositional variables that

describe what types of people the social context is composed of. For example, it is common to

ask respondents whether an alter they list is male or female, since that may be a key predictor

of a dependent variable. For each respondent this will result in a compositional variable

indicating the proportion of their network that is male and female. Continuous variables like age

will result in an average age of each respondent’s alters. The questions are repeated for each

alter listed. Ten questions about each alter, and 20 alters, will generate 200 questions. One

must balance the number of alters and the number of name interpreters. There are some things

respondents can be expected to know about their alters and others that are not reasonable to

ask. When selecting name interpreters try to use questions that respondents could answer

about all of their alters. Questions can often be modified from unreasonable to reasonable. For

example, a respondent may not know the specific age of their friend but could estimate it within

an age interval of ten years.

The last stage in the data collection process for an ego-centered network is the alter tie

evaluation which will result in a set of structural variables. In this stage respondents evaluate

whether their alters are tied in some way. The scale for the evaluation of these relations is more

narrow than with a whole network study since respondents are reporting on the way their

network members are tied, not on anything about themselves. It is typical to use a three point

scale for tie evaluation – the two alters are not tied, they might be tied, or they are definitely tied.

Alter tie evaluations are usually symmetric, meaning respondents are not asked to evaluate

whether a relationship between to alters is asymmetric. Since the ties are symmetric the

respondent only has to respond to (N*(N-1))/2 pairs. For 10 alters they must evaluate (10*(10-

1))/2 or 45 pairs. Figure 7 shows how the number of tie evaluations increases geometrically as

the number of alters increases.

Page 18: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

18

Figure 7. Number of alter pair evaluations as alters are added.

Some argue that alter tie evaluations may not be accurate citing studies showing that

respondents did not accurately report the interactions among members of a whole network to

which they belonged (Bernard et al. 1984). However there is a difference between reporting on

ties among co-members of a whole network, all of whom the respondent may not know, and

reporting on the members of one’s own personal network, whom respondents know, by

definition. Those who conduct ego-centered studies often contend that respondents can

accurately report about the interactions among members of their networks (see McCarty et al.

2007). And most social network analysts believe the tie evaluations at least result in cognitive

social structures; that is the social context the respondent believes exists, and that this has an

effect on attitudes and behaviors (Krackhardt 1990).

There are alternatives to having respondents evaluate each tie. McCarty and Killworth

(2007) showed that eliciting a larger set of alters, and then sampling at least 25 of those alters,

produces similar structural properties as asking for all tie evaluations among a larger set of

alters. Some programs, such as Vennmaker5, use a visual interface to collect network alters

and ties. The advantage of a visual interface is that respondents may find it to be a more

engaging format. With Vennmaker the respondent’s network is represented as a set of

concentric circles with the respondent in the center. The respondent drags icons representing

men and women into the appropriate spot and connects them using icons for strong and weak

ties. While the visual interface is appealing, it may introduce an experiment effect as

respondents connect mostly those who are grouped closely together.

Page 19: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

19

After data collection there are two ways in which personal network data are used. It is

typical to use a visualization of the personal network as a cue to interview the respondent about

the focus of the research (McCarty et al. 2007). This is illustrated with the personal network

visualization in Figure 8 depicting the network of a second generation West African in

Barcelona, Spain. The respondent lived in the same apartment with her Muslim family. The

nodes representing the network alters are labeled by the country of residence, colored by the

perceived skin color, sized by perceived closeness and shaped by whether or not they smoke

(squares are smokers and circles are non-smokers). Using these visualizations the authors

were able to ask the informant about the large number of isolates, and the unconnected dyad.

This, she said, accommodated her desire to smoke and to drink which was not acceptable to

others in her household. In other words, visualizations like these can be used as a platform to

talk to respondents about their lives.

Figure 8. Personal network visualizationsa of a second generation West African

The second way the data are used is to model the social network variables across

respondents. For example, Lubbers et al. (2007) used the compositional and structural

variables of personal networks to explain variability across respondents in ethnic identification.

They used a cluster analysis to classify migrants into five network profiles on the basis of eight

variables related to network composition (the percentage of Spanish alters, the percentage of

migrants, the percentage of family members, average closeness and average frequency of

contact with network members) and network structure (density, betweenness centralization, and

number of cohesive subgroups), as well as a variable that combined composition and structure

(homogeneity of cohesive subgroups). The resulting network profiles appeared to be

differentially related to ethnic identifications. In particular, migrants with so-called "dense family

networks" (networks in which network members—mostly family members and people from the

country of origin—formed one dense cluster) tended to exhibit ethnic- exclusive self-

identification, whereas migrants with more heterogeneous personal networks tended to exhibit

more plural definitions of belonging.

Combination of whole and ego-centered networks

Some research can benefit from both the whole-network and ego-centered approaches.

The benefit of the whole network approach is the detailed understanding of the social structure

Page 20: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

20

within a bounded group. The advantage of ego-centered networks is the inclusion of the social

context that cuts across many groups. A common way to do both is to conduct a whole network

study on, say, the members of a small village, and to use that network to identify respondents

for an ego-centered study. For example, one may want to know what resources, external to the

village, are available to those who are highly central within the whole network versus those who

are marginalized or isolated.

Another approach is combining a set of personal networks from respondents within a

boundary in an attempt to create a whole network. This approach has the advantage of

identifying critical nodes that may not be members of the bounded group, but are commonly

connected to it. For example, residents of a village may all mention an itinerant trader as an

alter in their personal network, while the trader is not actually a member of the village (Mathews

2010). The over-lapping network approach is useful for identifying broad structural properties of

a whole network, but (depending on the sample size) it cannot reliably represent fine-grained

properties.

Social Network Theory

There are several theoretical concepts that are used by social network analysts. Some

of them originated with the study of social networks; others originated elsewhere but are

frequently used in social network analysis

Social Capital

The concept of social capital, originally formulated by Bourdieu (1980, 1986) has been

developed by several authors as Lin (1995, 1999, 2001), Putnam (2000), and Portes (1998).

This latter author describes social capital as “…the ability of actors to secure benefits by virtue

of membership in social networks or other social structures”. The idea that social position may

give someone access to information or resources they can then use to their benefit is a useful

concept in many of the social sciences. Burt (2004) described the mechanisms by which

someone can use social position to their advantage. People who occupy brokering positions,

such as those with high betweenness centrality, are in a position to control the flow of

information and to synthesize information from different sources into something new. In this

sense social capital leads to the potential for creativity. A contrasting view is that of Krackhardt

(1999) whose notion of Simmelian ties is focused in part on conforming to conflicting norms.

When applying social networks to a project one should consider whether being in a brokering

position between two or more groups is an advantage, in terms of information control or a

burden in terms of the need to accommodate potentially competing interests. Some people may

occupy a brokering position but may not be aware of it or inclined to use it to their advantage.

Network closure refers to the tendency for a network to become interconnected, particularly in

small groups. Within small networks, such as a village, there may be strong pressures for

networks to become connected. Burt (2005) describes closure as an alternative form of social

capital given increased trust from the interconnectedness that comes from small groups.

Diffusion

Page 21: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

21

Diffusion of cultural traits used to be a controversial theory in anthropology, although

well-accepted at some levels by archaeologists. These theories focused at the societal level

looking for explanations of cultural traits distribution in wide geographical areas. Network

analysts typically examine the diffusion of innovations at the individual level, looking for changes

in behavior like the adoption of a given innovation (Rogers 2003). These ideas have been the

foundation for many public health interventions (see Valente and Pumpuang, 2007).

Strong and Weak Ties

Grannovetter (1973) introduced the notion that people have strong ties to some people

and weak ties to others. He showed that for some tasks, such as looking for a job, the weak tie

may be better than the strong tie because the weak tie is a link to new opportunities. Strong ties

tend to be redundant and connected, such as family and very close friends. These strong ties

not only tend recycle the same information, they may compete for the same opportunities.

Weak ties can be thought of as brokerage across social situations.

Small World

In 1967 Stanley Milgram wrote an article for the first issue of Psychology Today about an

experiment he conducted trying to measure the average number of links between any two

randomly selected people in the U.S. Most people are now familiar with the notion that any two

people are linked on average by less than six intermediaries (six degrees of separation).

Killworth et al. (1984) examined the nature of the choices people made in choosing

intermediaries, showing that occupation and location explained most of these choices. In 1998

Watts and Strogatz showed mathematically that a relatively small number of non-random

connections, such as through occupation and location, explains the short chains.

Anthropologists may use the small world concept in understanding short and long connections

between members of cultural groups.

Scale-free networks

Following the article by Watts and Strogatz came a surge of interest by physicists in

social network problems. As the internet led to social media sites such as Mypace and

Facebook and a proliferation of web sites, the approach of physicists that lends itself to

modeling of large amounts of data became a unique thread of research in the field of social

networks. Borgatti et al (2009) outline the difference between the approach of social scientists

and that of physicists.

Scale free networks refers to the idea that the basic structure of a network is very much

the same for both small and large networks; in other words the structure is not determined by

the size of the network. Some types of networks, such as web site connections, are scale free.

While scale free networks were first observed in citation networks, physicists have focused on

them since the late1990s. Anthropologists may find that some social phenomena they research

are scale-free.

Social Learning

Page 22: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

22

Social leaning theory (Bandura 1977) is simply the idea that people learn from their

social context; that is they learn from others. The notion of social context is easily

operationalized using social networks, both whole and ego-centered. This theory lends itself to

studies that focus on information transmission and the social avenues and impediments to that

transmission.

Potential Issues with the Institutional Review Board (IRB)

Although social network analysis has been an established discipline for more than 80

years, the advent of social networking sites has dramatically increased the visibility and

application of the social network approach. Each year more disciplines and journals are

publishing articles using social networks as a way to operationalize concepts. For decades

IRBs approved many projects that they are now reconsidering (Borgatti and Molina 2005).

Across universities the IRBs are not necessarily consistent in these evaluations.

Perhaps the most important IRB-related issue as of this writing is the controversy over

the rights of network members mentioned but not interviewed. This is far more of an issue with

ego-centered network studies than it is with whole network studies. This issue becomes

particularly problematic when the study involves behaviors or conditions that are illegal or

stigmatizing. For example, a social network study of migrants where respondents are asked to

indicate whether their network alters were legal or not legal migrants would not be approved by

some IRBs or research panels. The assumption is that those migrants whose immigration

status was reported did not have an opportunity to consent and therefore their rights were

violated. This is less of an issue with whole networks because the likelihood is higher that

anyone whose behavior is reported is also a respondent. However it is possible that a member

of a whole network could fail to respond and then the same rules would potentially apply.

One could, of course, avoid asking any question that involves illegal or stigmatizing

behavior. However that would preclude using a network approach to understand controversial

social problems. When possible researchers can try to avoid real names, although some

network techniques require them, particularly whole network studies. Some IRBs will be

satisfied with more detailed consent forms or careful data encryption and data handling after

data have been collected.

Limitations of Social Networks

Studies using a social network approach or studying social networks have proliferated in

the last decade. As of this writing, the Web of Science had 6,270 references to articles that

contain the term social network. Since 2003 the number of articles has grown exponentially,

with just over 200 in 2003 and over a 1,000 in 2012. These references are spread across many

disciplines, with the highest number from public health (677), sociology (508), psychiatry (483)

and management (388). Anthropology had 208. Assuming the trend continues more

researchers will apply social network analysis as a method or framework for collecting and

analyzing their data.

Page 23: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

23

But social network analysis is not a panacea. Researchers are often attracted to it

because it is trendy and because network visualizations tend to be compelling. Although it can

be an appropriate approach, there is the danger that a relatively straightforward concept may be

over-complicated. Boissevain (1979) put this poignantly:

“The battery of techniques with which social scientists have equipped themselves to answer the

limited questions that network analysis can resolve produces overkill. Flies are killed with

dynamite.” (Pg. 393)

While social network analysis is hardly dynamite, there are several reasons one should

think twice before incorporating it into a research design.

First, social network data are typically difficult to collect. One-mode, whole network data

that rely on a survey typically involve high researcher burden as tie evaluations must be

collected from as many members of the group as possible. When the group is relatively small

and easily located this is feasible. But with large groups, or groups that are dispersed, the

logistics can get out of hand very quickly. With large networks it becomes increasingly likely

that some data will be missing. In contrast to whole networks, ego-centered network data tend

to be high on respondent burden as respondents must answer many questions about the

network alters they mention.

Once network data are collected they can be difficult to manage. While software

packages such as UCInet and Egonet have made network data easier to work with, even these

programs require a certain amount of expertise. In the case of whole network analysis

packages such as UCInet, the data are sometimes not in a form that can be input without some

manipulation. Like any type of data, quantitative or qualitative, one must be prepared for the

commitment to learning the details and the peculiarities of social network data.

Although readers and reviewers in some disciplines are increasingly becoming well-

informed about network analysis, most anthropologists are not familiar with it. This means that

until network analysis becomes an established method in anthropology considerable effort must

be made to explain the data and analysis to reviewers and readers. The network visualizations

that may attract users in the first place are often confusing to those who do not understand

them. What may seem a random collection of dots and lines could come out in a variety of

ways. For example, a network that is highly cohesive may lack disconnected groups or isolates.

The contrast of cohesion versus disconnectedness may be a basis for explanation.

Lastly, it is not uncommon for researchers to try to apply the network approach when

there is no network. It is possible to make a network out of many types of data that do not

comprise an interacting set of actors. For example one could use a two-mode approach to

record different types of foods that a set of respondents eat. Network software can be used to

create a network of the actors where the tie represents the foods they eat in common. Eating

similar foods does not imply interaction.

Despite these limitations, social network analysis has proven to be a useful tool in the

social sciences. Anthropologists were among the early developers of network analysis and the

Page 24: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

24

methods and theories of interaction that comprise network analysis are increasingly being used

again by anthropologists in creative ways. Used correctly social network analysis captures

something unique – the pattern of social relations. As the social network approach is applied in

other disciplines it will also establish a common set of variables for collaboration with

researchers in both the natural and social sciences.

Page 25: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

25

References

Adler-Lomnitz, Larissa (1975). Cómo sobreviven los marginados. Madrid: Siglo XXI.

Aral, Sinan, Lev Muchnika, and Arun Sundararajana (2009). Distinguishing influence-based

contagion from homophily-driven diffusion in dynamic networks, PNAS December 22, Vol. 106

(51): 21544-21549.

Bandura, Albert (1977). Social Learning Theory. Englewood Cliffs, N.J. : Prentice Hall.

Barbosa de Almeida and Mauro William (2010). On the Structure of Dravidian Structure of

Dravidian Relationship Systems, Mathematical Anthropology And Cultural Theory: An

International Journal, Volume 3 no. 1 August [www.mathematicalanthropology.org.

Barnes, John (1954) "Class and committees in a Norwegian Islan Parish", Human Relations,

vol. 7, núm 1; 39-58.

Barnett, G. A., J. A. Danowski, T. H. Feeley, and J. Stalker (2010). Measuring quality in

communication doctoral education using network analysis of faculty-hiring patterns. Journal of

Communication 60:388–411.

Berkman, L. F., and S. L. Syme (1979). Social networks, host resistance, and mortality: A nine-

year follow-up study of alameda county residents. American Journal of Epidemiology, 109(2),

186-204.

Bernard, H. Russell; Killworth, Peter D.; Kronenfeld, David; Sailer, Lee (1984) “The Problem Of

Informant Accuracy - The Validity Of Retrospective Data” Annual Review Of Anthropology 13 :

495-517 .

Bernard, Russell H., Peter D. Killworth, Michael J. Evans, Christopher McCarty and Gene Ann

Shelley (1988). Studying social relations cross-culturally, Ethnology, Vol 27, 2 (155-179).

Bernard, Russell H., Eugene C. Johnsen, Peter D. Killworth, Christopher McCarty, Gene A.

Shelley and Scott Robinson (1990). Comparing four different methods for measuring personal

social networks, Social Networks 12:179-215.

Bernard, H Russell, Hallett, Tim, Iovita, Alexandrina, Johnsen, Eugene C, Lyerla, Rob ,

McCarty, Christopher , Mahy, Mary , Salganik, Matthew J , Saliuk, Tetiana ,Scutelniciuc, Otilia,

Shelley, Gene A , Sirinirund, Petchsri ,Weir, Sharon, Stroup. Donna F (2010). Counting hard-to-

count populations: the network scale-up method for public health, Sexually Transmitted

Infections, 86: ii11-ii15.

Boissevain, Jeremy (1974). Friends of Friends: Networks, Manipulators and Coalitions. London:

Basil Blackwell.

Boissevain, Jeremy (1979) Network Analysis: A Reappraisal Current Anthropology, 20 (2): 392-

394.

Page 26: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

26

Boorman, Scott A., and Harrison C. White (1976). Social Structure from Multiple Networks, II:

Role Structure, American Journal of Sociology, 81, pp 1384-1446.

Borgatti, Stephen. P. (2002). NetDraw Software for Network Visualization. Analytic

Technologies: Lexington, KY.

Borgatti, Stephen P., Everett, M.G. and Freeman, L.C. (2002). Ucinet for Windows: Software for

Social Network Analysis. Harvard, MA: Analytic Technologies.

Borgatti, Stephen P. and Jose-Luis Molina (2002). Toward Ethical Guidelines for Network Research in Organizations. Social Networks 27(2): 107-117. Borgatti, Stephen P. ; Mehra, Ajay; Brass, Daniel; Labianca, J. Giuseppe (2009). Network

Analysis in the Social Sciences, Science 323: 892-895

Borgatti, Stephen P., Martin G. Everett and Jeffrey C. Johnson (2013), Analyzing Social Networks. Sage: in press. Bott, Elizabet (1957). Family and social network. London: Tavistock Institute of Human

Relations.

Bourdieu, P. (1980), ‘Le capital social: notes provisoires’, Actes de la Recherche en Sciences

Sociales, 3, 2-3.

Bourdieu, P. (1986), ‘The forms of capital’, in J.G. Richardson (ed.), Handbook of Theory and

Research for the Sociology of Education. New York: Greenwood Press, pp. 241-258.

Burt, Ron (2004). Structural Holes and Good Ideas. American Journal of Sociology 110: 349–

399

Burt, Ron (2005) Brokerage and Closure : An Introduction to Social Capital. Oxford ; New York:

Oxford University Press.

Berkes, Fikret; Folke, Carl (eds.) (1998). Linking Social and Ecological Systems. Management

Practices and Social Mechanisms for Building Resilience. Cambridge: Cambridge University

Press.

Calvet, Laura; Molina, JL and Victoria Reyes-García (2012) Seed exchange as an

agrobiodiversity conservation mechanism. A case study in Vall Fosca, Catalan Pyrenees,

Iberian Peninsula, Ecology and Society 17 (1): 29.

Cohen, S., and T.A. Wills (1985). Stress, social support, and the buffering hypothesis.

Psychological Bulletin, 98(2), 310-357.

Coleman, J.S. (1961). The adolescent society: The social life of the teenager and its impact on

education. New York: Free Press of Glencoe.

Coleman, J. S., E. Katz and H. Menzel. 1966. Medical innovation. A diffussion study.

Indianapolis: The Bobbs-Merril Company.

Page 27: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

27

Connelly, Brian L.; Johnson, Jonathan L.; Tihanyi, Laszlo; Ellstrand, Alan E. (2011) More Than

Adopters: Competing Influences in the Interlocking Directorate. Organization Science. 22(3):

688–703

Costenbader, Elizabeth; Valente, Thomas W. (2003) The stability of centrality measures when

networks are sampled. Social Networks 25: 283–307.

Christakis, MD & James H. Fowler (2009) Connected. The Surprising Power of Our Social

Netwroks and How They Shape Our Lives. New York: Litte, Brown and Company.

de Nooy, Wouter; Mrvar, Andrej & Vladimir Batagelj (2005). Exploratory Social Network Analysis

with Pajek. Cambridge: Cambridge University Press.

Dunbar, Robin (2008) Cognitive constraints on the structure and dynamics of social networks.

Group Dynamics-Theory Research and Practice. 12(1): 7-16

Ellen, Roy, Peter Parkes and Alan Bicker (2000) Indigenous Environmental Knowledge and its

Transformations. Critical Anthropological Perspectives. Amsterdam: Harwood Academic

Publishers.

Epstein, A. L. (1969) "Gossip, Norms and Social Networks". In: Mitchell, Clyde J., Social

Networks in Urban Situations. Analyses of Personal Relationships in Central African Towns.

Manchester: Manchester University Press.

Feld, Scott (1981) The Focused Organization of Social Ties. American Journal of Sociology,

86(5): 1015-1035.

Fischer, Claude S. (1982). To dwell among friends: Personal networks in town and city.

Chicago: University of Chicago Press.

Freeman, Linton C. (1999) Visualizing social networks. In

ttp://www2.heinz.cmu.edu/project/INSNA/joss/vsn.html

Freeman, L. C. (2004). The Development of Social Network Analysis: A Study in the Sociology

of Science. Empirical Press.

Gonçalves B, Perra N, Vespignani A, (2011). Modeling Users' Activity on Twitter Networks:

Validation of Dunbar's Number. PLoS ONE 6(8): e22656. doi:10.1371/journal.pone.0022656.

Granovetter, Mark (1973) The strength of weak ties, American Journal of Sociology 78-1361-

1381.

Hamberger, K. (2011). Matrimonial circuits in kinship networks: Calculation, enumeration and

census, Social Networks, 33(2), 113-128.

Hanneman, Robert A. and Mark Riddle (2005). Introduction to social network methods.

Riverside, CA: University of California, Riverside (published in digital form at

http://faculty.ucr.edu/~hanneman/).

Page 28: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

28

Johnsen, E. C. (1985). Network macrostructure models for the davis-leinhardt set of empirical

sociomatrices, Social Networks, 7(3), 203-224.

Kadushin, Charles (1982). Social Density and Mental Health. In: Marsden, Peter V. and N. Lin,

Social Structure and Network Analysis. Beverly Hills / London / New Delhi: Sage Publications,

pp.147-158.

Kadushin, Charles (1968). Power, Influence and Social Circles: A New Methodology for

Studying Opinion Makers, American Sociological Review, Vol. 33, No. 5. pp. 685-699.

Kadushin, Charles (2012). Understanding Social Networks: Theories, Concepts and Findings.

New York: Oxford University Press.

Kapferer, Bruce (1972). Strategy and transaction in an African factory: African workers and

Indian management in a Zambian town. Manchester: Manchester University Press.

Kasper, C., Voelkl, B. (2009). A social network analysis of primate groups, Primates, 50 (4):

343-356.

Killworth, P. D., & Bernard, H. R. (1979). Informant accuracy in social network data III: A

comparison of triadic structure in behavioral and cognitive data. Social Networks, 2(1), 19-46.

Klovdahl, A. S. (1985). Social networks and the spread of infectious diseases: The AIDS

example. Social Science and Medicine, 21(11), 1203-1216.

Krackhardt, David (1990). Assessing the Political Landscape: Structure, Cognition, and Power

in Organizations, Administrative Science Quarterly, 35 pp. 342-69.

Laumann, Ed (1973). The Bonds of Pluralism: The Forms and Substance of Urban Social

Networks. New York: Wiley.

Latkin, C. A., Forman, V., Knowlton, A., and S. Sherman (2003). Norms, social networks, and

HIV-related risk behaviors among urban disadvantaged drug users. Social Science and

Medicine, 56(3), 465-476.

Lorrain, F. and White, H.C. (1971). Structural Equivalence of Individuals in Social Networks,

Journal of Mathematical Sociology, 1(1): 49-80.

Mitchell, Clyde J. (1972). Foreword. In: Kapferer, Bruce (1972). Strategy and transaction in an

African factory: African workers and Indian management in a Zambian town. Manchester:

Manchester University Press.

Hage, Per and Frank Harary (1997). Island Networks: Communication, Kinship and

Classification Structures in Oceania. , Cambridge: Cambridge University Press..

Hanneman, Bob and Mark Riddle, http://faculty.ucr.edu/~hanneman/nettext/

Page 29: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

29

Hopkins, Allison (2011). Use of Network Centrality Measures to Explain Individual Levels of

Herbal Remedy Cultural Competence among the Yucatec Maya in Tabi, Mexico. Field Methods

23(3): 307-328

Houseman, M., and D. R. White (1996). Structures réticulaires de la pratique matrimoniale,

L’Homme 139, 59–85.

Jackson, Matthew O. (2008) Social and Economic Networks. Princenton, NJ : Princeton

University Press.

Johnson, Jeffrey C. (1994). Anthropological Contributions to the Study of Social Networks: A

Review, In: Wasserman, S. and J. Galaskiewicz, eds., Advances in Social Network Analysis.

Research in the Social and Behavioral Sciences, Thousand Oaks, CA: Sage.

Kasper, Claudia; Voelkl, Bernhard (2009). A social network analysis of primate groups.

Primates. 50(4): 343-356.

Killworth, P.D. & Bernard, H. R (1976). Informant accuracy in social network data, Human

Organization, 35: 269-286.

Killworth, Peter D.; Bernard, H. Russell and Christopher McCarty (1984). Measuring Patterns of

Acquaintance, Current Anthropology 25:381-397.

Kim, S. and E.H. Shin (2002). A longitudinal analysis of globalization and regionalization in

international trade: A social network approach, Social Forces 81(2): 445-471

Krackhardt, David (1999) The ties that torture: Simmelian tie analysis in organizations.

Research in the Sociology of Organizations 16: 183-210.

Lin, Nan (1995). Social resources: A theory of social capital. Revue francaise de sociologie,

36(4): 685-704.

Lin, Nan (1999). Social Networks and Status Attainment, Annual Review of Sociology, 25, 467-

487.

Lin, Nan (2001). Social Capital: A Theory of Social Structure and Action. Cambridge. Cambridge

University Press.

Lubbers, Miranda ; Molina, José Luis; McCarty, Christopher (2007) Personal Networks and

Ethnic Identifications: The Case of Migrants in Spain, International Sociology 22:720-740

Martin, CL; Fabes, RA; Hanish, LD; Holleristein, T (2005) Social dynamics in the preschool.

Developmental Review 25(3-4): 299-327.

Mathews, Mason Clay (2010). Socio-economic change in the transition from patron-client to

social movement networks in Brazilian Amazonia. University of Florida, ProQuest, UMI

Dissertations Publishing, 3467581.

Page 30: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

30

McCarty, Christopher; Killworth, Peter (2007) Impact of Methods for Reducing Respondent

Burden on Personal Network Structural Measures Social Networks Social Networks 29: 300-

315.

McCarty, Christopher; Molina, José Luis; Aguilar, Claudia y Laura Rota (2007). A Comparison of

Social Network Mapping and Personal Network Visualization, Field Methods, 19 (2) 145-162.

McCarty, Christopher; Molina, Jose Luis; Aguilar, Claudia; Rota, Laura (2007) A comparison of

social network mapping and personal network visualization. Field Methods 19: 145-162.

Milgram, Stanley (1967) The small world problem, Psychology Today, 1 (1): 61‐67.

Mitchell, Clyde J. (1969) Social Networks in Urban Situations. Analyses of Personal

Relationships in Central African Towns. Manchester: Manchester University Press.

Molina, Jose Luis; Maya-Jariego, Isidro; McCarty, Christopher (forthcoming) Giving Meaning to

Social Networks: Methodology for Conducting and Analyzing Interviews based on Personal

Network Visualizations. In: Holstein, B. and S. Dominguez, eds., Mixed Methods Network

Research, Cambridge: Cambridge University Press.

Molina JL, Petermann S. and A. Hertz (2012). “Defining and Measuring Transnational Fields”,

MMG Working Papers Print. 16.

Moreno, Jacob L. (1934) Who Shall Survive? Foundations of Sociometry. New York: Beacon

House.

Mouttapa, M; Valente, T; Gallaher, P; Rohrbach, LA; Unger, JB (2004) Social network

predictors of bullying and victimization. Adolescence. 39(154): 315-335

Nadel, S.F. (1957) Teoria de la estructura social [Theory of social Structure. 1957]. Madrid:

Guadarrama,

Onnela, JP Saramaki, J; Hyvonen, J ; Szabo, G; Lazer, D; Kaski, K; Kertesz, J ; Barabasi, AL

(2007) Structure and tie strengths in mobile communication networks. Proceedings of the

National Academy of Sciences of the United States of America. 104(18): 7332-7336

Padgett JF Ansell CK (2003) Robust Action And The Rise Of The Medici, 1400-1434. American

Journal of Sociology 98(6): 1259-1319

Portes. Alejandro (1998) Social Capital: Its Origins and Applications in Modern Sociology,

Annual Review of Sociology 1998. 24:1–24.

Putnam, R.D. (2000). Bowling Alone: The Collapse and Revival of American Community. New

York: Simon & Schuster.

Reyes-García V., Molina JL, Broesch J., Calvet L, Huanca T., Leonard WR, McDade TW., Saus

J, Tanner S. (2008) Do the aged and knowledgeable men enjoy more prestige? A test of

predictions from the prestige-bias model of cultural transmission, Evolution and Human

Behavior 29 (275-281).

Page 31: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

31

Reyes-García, Victoria; José Luis Molina, Thomas W. McDade, Susan Tanner, Tomás Huanca,

William R. Leonard, TAPS Bolivian Research Team (2009) Inequality in social rank and

nutritional status. Evidence from a pre-industrial society in the Bolivian Amazon, Social Science

and Medicine, 69 (571-578).

Rice, R.E. (1994). Relating electronic mail use and network structure to R&D work networks and

performance. Journal of Management Information Systems, 11:9–20.

Richards, W. Jr. and G. A. Barnett. eds. (1993). Progress in Communication Sciences:

Advances in communication network analysis. Norwood, NJ: Ablex.

Robins, G.L., Snijders, T.A.B., and P. Wang (2005). Recent developments in exponential

random graph (p*) models for social networks. Social Networks.

Rogers, E. M. (2003). Diffusion of Innovations. 5th ed. New York: Free Press.

Rohilla Shalizi, Cosma and Andrew C. Thomas (2011). Homophily and Contagion Are

Generically Confounded in Observational Social Network Studies, Sociological Methods &

Research, 40: 211-239.

Scott John (2000). Network Analysis: A Handbook. (2nd Edition) Newbury Park CA: Sage.

Schweizer, Thomas (1997). Embeddedness of Etnographic Cases: A Social Networks

Perspective", Current Anthropology, vol. 38 (5), pp. 739-760.

Snijders, Tom A.B. (2005). Models for Longitudinal Network Data. In: P. Carrington, J. Scott, &

S. Wasserman (eds.), Models and methods in social network analysis. New York: Cambridge

University Press, pp. 215-247.

Snijders, T.A.B., van de Bunt, G.G., and Steglich, C.E.G. (2010). Introduction to actor-based

models for network dynamics, Social Networks, 32, 44-60.

Srinivas M.N. and A. Béteille (1964). Networks in Indian Social Structure, Man 64(212).

Valente, Thomas (1997) Social Network Associations with Contraceptive Use Among

Cameroonian Women in Voluntary Associations, Social Science & Medicine, 45(5): 677-87.

Valente, Thomas W.; Pumpuang, Patchareeya (2007) Identifying opinion leaders to promote

behavior change. Health Education & Behavior 34(6): 881-896

Valente, Thomas (2010). Social Networks and Health: Models, Methods and Applications.

Oxford: Oxford University Press.

VanderWeele, Tyler J. (2011). Sensitivity Analysis for Contagion Effects in Social Networks

Sociological Methods & Research, 40: 240-255.

Vertovec, Steven (2009). Transnationalism. Series: Key Ideas. New York: Routledge.

Page 32: PREPRINT - pagines.uab.catpagines.uab.cat/joseluismolina/sites/pagines.uab.cat.joseluismolina... · PREPRINT McCarty, C & JL Molina. “Social Network Analysis”. In: Bernard, H.R

32

Wellman, Barry (1979). The Community Question: The Intimate Networks of East Yorkers.

American Journal of Sociology, 84(5): 1201-1231.

Wasserman, Stanley and Faust Katherine (1994). Social Network Analysis: Methods and

Applications. Cambridge: Cambridge University Press.

Wasserman, S. and Pattison, P. E. (1996). Logit models and logistic regressions for social

networks: I. An introduction to Markov graphs and p*, Psychometrika, 61, 401-425.

Wasserman, S., and G. Robins (2005). An Introduction to Random Graphs, Dependence

Graphs, and p*. In Carrington, Scott & Wasserman (Eds.) Models and Methods in Social

Network Analysis. New York: Cambridge University Press,

Watts, Duncan J.; Strogatz, Steven H. (1998) Collective dynamics of 'small-world' networks.

Nature 393 (6684): 440–442.

White, D. R., Batagelj, V., & Mrvar, A. (1999). Anthropology: Analyzing large kinship and

marriage networks with pgraph and pajek. Social Science Computer Review, 17(3), 245-274

White, Douglas R. and Ulla C. Johansen (2005). Network Analysis and Ethnographic Problems.

Process Models of a Turkish Nomad Clan. Plymouth: Lexington Books.

White, Harrison C., Scott A. Boorman, and Ronald L. Breiger (1976). Social Structure from

Multiple Networks, I: Blockmodels of Roles and Positions, American Journal of Sociology, 81,

pp. 730-780.

Wolfe, Alvin W. (1978) The rise of network thinking in anthropology, Social Networks Volume 1,

Issue 1 (53-64).

1 Ucinet is a program developed by Steve Borgatti, Martin Everett and Lin Freeman and sold through the website

www.analytictech.com. As of this writing the price is $40 for students and $150 for faculty and government agencies. Pajek is a free program available at http://vlado.fmf.uni-lj.si/pub/networks/pajek/. Ucinet tends to be easier to learn than Pajek. Pajek is more efficient in analyzing large datasets and developing network models. 2 Netdraw is a free program for social network visualization available from www.analytictech.com. It is included

with the program Ucinet. There is a tutorial to learn social network analysis using Netdraw at http://faculty.ucr.edu/~hanneman/nettext/. 3 http://www.stats.ox.ac.uk/~snijders/siena/

4 http://sourceforge.net/projects/egonet/

5 http://www.vennmaker.com/en/