www.helsinki.fi/crc

Measuring the effect of social connections on political activity on

FacebookInternet, Politics, Policy 2012: Big Data, Big Challenges

Oxford, UK, Sep 20. 2012

Olli Parviainen, Petro Poutanen, Salla-Maaria Laaksonen & Mikael RekolaCommunications Research Centre CRC / University of Helsinki

Faculty of Social Sciences / Department of Social Research / Media & Communication Studies

Outline1. Introduction2. Data, methods & variables3. Research questions4. Comparing support groups5. Comparing user and admin initiated

communication6. Discussion

Introduction

Basic details The study examines online political behavior

in Facebook Network analysis and statistical analysis are

used Case: Second round of the Finnish

presidential elections 2012 Massive campaigning on Facebook Comparative study of the two supporter

populations

Data, methods & variables

Data and method Data extracted post hoc from Facebook platform via it's

FQL2 interface Collected and analyzed using C++, Perl, Graph.pm, Gephi

and SPSS Data comprises FB pages’ activities (wallposts, comments,

wallpost likes, comment likes) and structures (friendship connections)

Social network analysis (Wasserman & Faust, 1994; Monge & Contractor, 2003) and traditional statistical methods (time series, correlation, and regression analysis)

Likes The number of likes the page has in the time of the post.

Number of wall post likes The number of likes the wall post has received

Number of comments The number of comments posted to the wall post

Number of comment likes The number of likes the comments within the wall post have received

Overall activity Sum of all activity (wall post likes, comments and comment likes). Measures the response for the post.

Active users Absolute number of different users activated in the post.

Activity level The share of the active users of page's all likers in the wall post . Measures the wall post's ability to engage the page likers (audience)

Number of wall post likers The number of different users liking the wall post

Number of commenters The number of different users commenting on the wall post

Number of comment likers The number of different users liking comments

Number of components Absolute number of friendship components within the post

Friendship network edges The number of friendship connections within the post

Friendship average component size

Mean of all friendship component sizes within the post

Friend average degree Mean of number of friends the active users of the post have each other

Friend overall degree Mean of number of friends the active users of the post have with all the active users in the two week time frame

Friend clustering coefficient The clustering coefficient of the friendships

Network friends percentage The percentage of the active users of the post who have at least one friend among the other active users

Poster friend count Number of friends the author of the post has within all the active users of the page

Research questions

What kinds of friendship structures are typical in large support groups (e.g. dyad, triads, bigger cliques, communities)?

How do people act in support pages (likes, comment likes, comments, wall posts)?

How activities are associated with the friendship structures of the support pages?

How the interaction patterns are associated with the friendship structures of the support pages?

Comparing two support pages

Count of likes and posts

Overall activities & active users

Network structure shows more clustering on the Niinistö page

Nodes Edges Diameter Radius Avg. path lg. Avg. degree central. Avg. clust. coeff.Niinistö 35372 189673 18 9 4.699 10.72 .1374Haavisto 57696 461744 13 7 6.804 16.01 .1129

Comparing admin and user initiated posts

Results:

Niinistö Haavisto050000

100000150000200000250000300000350000400000

Overall activity (count of all activities)

UserAdmin

Niinistö HaavistoUser 9435 17818Admin 84 127

Regression coefficients explaining user generated activity level

Number of components within the post

Avg. Friendship centrality degree within the post

Niinistö 1,075 0,511Haavisto 0,999 0,8

All coeffiecients are highly statistically significant

In admin-initiated posts users’ intra-post connedtedness is associated with bigger

activity in Niinistö page

Discussion

Conclusion

Niinistö Haavisto

Activity More admin initiated

More user initiated

Structure Cliques, wide Dense

Interaction More friendship based (”friends

interacting”)

More community based (”strangers

interacting)

Implications and challenges Practical implication

enhancing means for political campaining and public relations practice Scientific implications

Gaining more (accurate) information on social behaviour in online social networks

Methodological contribution: SNA & statistics & large real world data sets Challenges

The platform infrastrucutre determines the activity heavily. For example, how to identify the effects of the technology and include it in the analysis, for example FB Edgerank?

Content of the posts matters: combining textual content analysis with activity and network measures is needed

Contentual factors: external events, news media, gallups Privacy issues: demographic variables are difficult to incorporate

Thank you!

olli.parviainen@helsinki.fi

petro.poutanen@helsinki.fi

salla.laaksonen@helsinki.fi

mikael.rekola@helsinki.fi