Source-linked AI summary
Partisan Asymmetries in Online Political Activity
Michael D. Conover, Bruno Gonçalves, Alessandro Flammini, Filippo Menczer
TL;DR
The paper asks whether partisan engagement in American online politics has shifted after the Democratic advantage reported for 2008. It analyzes Twitter behavior and networks among politically active users around the 2010 midterms, finding greater activity, denser social connectivity, and more rapidly diffusive political communication among right-leaning users.
Problem
The paper addresses limited evidence about changing partisan asymmetries in online political engagement after the widely reported Democratic online-mobilization advantage in 2008.
Method
The authors analyze behavior, communication, and social networks of more than 18,000 politically active Twitter users using partisan cluster membership as a high-fidelity affiliation proxy.
Results
Right-leaning users produce more than 50% more political content, show tighter social connectivity, and participate in networks facilitating rapid and broad political-information dissemination.
Takeaways & Limitations
The findings indicate a shifting landscape in which right-leaning Twitter users form a highly active and densely interconnected political constituency.
Abstract
from arXiv · showhide
We examine partisan differences in the behavior, communication patterns and social interactions of more than 18,000 politically-active Twitter users to produce evidence that points to changing levels of partisan engagement with the American online political landscape. Analysis of a network defined by the communication activity of these users in proximity to the 2010 midterm congressional elections reveals a highly segregated, well clustered partisan community structure. Using cluster membership as a high-fidelity (87% accuracy) proxy for political affiliation, we characterize a wide range of differences in the behavior, communication and social connectivity of left- and right-leaning Twitter users. We find that in contrast to the online political dynamics of the 2008 campaign, right-leaning Twitter users exhibit greater levels of political activity, a more tightly interconnected social structure, and a communication network topology that facilitates the rapid and broad dissemination of political information.
I. INTRODUCTION
The paper investigates partisan asymmetries in online political engagement after earlier evidence of a Democratic advantage in 2008. Examining more than 18,000 politically active Twitter users, it finds greater activity, tighter social bonds, and more information-diffusive network structure among right-leaning users.
- Motivation: 24% of American adults got most 2010 midterm election news online, while 54% went online for political information.Among internet users who voted, 35% reported that online political information influenced their vote choice.
- Motivation: 2008 survey evidence found Obama voters more likely than McCain voters to create and share political content and engage politically on social networks.
- Research aim: The study examines partisan differences in behavior, communication patterns, and social interactions among more than 18,000 politically active Twitter users.Twitter is studied because its content is public, API-accessible, oriented toward information sharing, and prominent in American political discourse.
- Main findings: Right-leaning users exhibit greater political activity, tighter social bonds, and network topology facilitating rapid and broad political-information dissemination.This contrasts with the online political dynamics reported for the 2008 campaign.
- Individual behavior: Right-leaning users produce more than 50% more political content and devote a greater share of time to political discourse.They are also more likely to use hyperlinks and nearly twice as likely to self-identify politically in profile biographies.
- Connectivity: Right-leaning users show more mutually affirmed ties and connections with more individuals, while their retweet structure supports rapid and broad information dissemination.
II. PLATFORM & DATA
The study treats Twitter as a platform for broadcasting, following, direct interaction, and audience expansion through hashtags. It includes political hashtags related to progressive and conservative seed tags.
- Twitter features: Twitter users post 140-character messages called tweets and interact through follows, retweets, mentions, and hashtags.
- Follow relationships: Following is a directed, non-reciprocal link through which users subscribe to another user’s content stream.Users can also sample tweets containing specified hashtags.
- Hashtags: Hashtags identify tweet topics or intended audiences and can broaden a message’s potential audience beyond immediate followers.The political hashtags #tcot and #p2 denote “Top Conservatives on Twitter” and “Progressives 2.0.”
- Direct interaction: Retweets rebroadcast other users’ content and often act as endorsements, while mentions address or refer to users through the public feed.
- Political hashtags: Tweets containing political hashtags related to #p2 and #tcot were included in the sample.
B. Data
The data comprise an eighteen-week Twitter sample around the 2010 midterm elections, classified using political hashtags and prior partisan cluster labels. Hashtag selection and representativeness checks were used to define a focused political corpus.
- Data collection: The dataset covers September 1, 2010 through January 7, 2011, an eighteen-week period surrounding the November 4 congressional midterm elections.Twitter’s gardenhose API supplied approximately 10% of the full Twitter corpus.
- Data collection: 6,747 right-leaning and 10,741 left-leaning users produced 1,390,528 and 2,420,370 tweets, respectively.All gardenhose tweets from each user were evaluated to compare the proportions of attention devoted to political communication.
- Classification: The analysis uses political hashtags and partisan cluster-membership labels established in an earlier study of political polarization.The authors also assess whether the resulting networks and communities represent domestic political communication on Twitter generally.
- Identifying political content: Political communication is defined as any tweet containing at least one political hashtag.This restriction creates a high-fidelity corpus and avoids noise from topic-detection strategies, although hashtag-free political communication may be excluded.
- Hashtag discovery: Hashtag discovery seeded #p2 and #tcot, ranked co-occurring tags by the Jaccard coefficient, and identified 66 unique hashtags.Eleven hashtags were excluded because their meanings were overly broad or ambiguous.
- Representativeness: Figure 1’s roughly exponential popularity decay indicates that adding more political hashtags is unlikely to substantially increase corpus size or alter its structure.
B. Representativeness
The hashtag-based corpus captures a substantial and representative share of political communication, while additional hashtags yield diminishing gains in unique users and tweets.
- A sharp decay in hashtag popularity indicates that adding further political hashtags is unlikely to substantially increase or alter the corpus.
- Diminishing returns arise because many tweets and users are redundantly associated with multiple hashtags.
- Figure 2 orders hashtags by associated tweet totals and user totals while tracking the resulting unique users and tweets.
- 26.4% of 312,560 hashtagged tweets containing 2,500 election-related political keywords were covered by the target hashtags.
- Among the ten most popular excluded hashtags, only one was explicitly political and represented less than 2% of public political communication.
C. Inferring Political Identities from Communication Networks
The study infers political identities from partisan communities in a political retweet network, using clustering and manual content annotation to validate the proxy.
- The political retweet network represents information propagation from an original broadcaster to a user who retweets that content.
- Network clustering divides users in the largest connected component into two distinct communities.
- Cluster assignments combine label propagation seeded by modularity maximization, with assignments robust to starting-condition fluctuations.
- Community membership is used as a proxy for political identities across 18,470 users in the largest connected component.
IV. BEHAVIOR: INDIVIDUAL-LEVEL POLITICAL ACTIVITY
The analysis finds greater individual-level political engagement among right-leaning Twitter users, despite broadly comparable overall tweeting levels between the groups.
- Right-leaning users produce less total political content than left-leaning users while producing approximately the same number of tweets per user.
- Right-leaning users allocate proportionally more activity to political communication and are more likely to reveal ideology in biographies and share hyperlinks.
- 54% more total political content is produced by right-leaning users despite their comprising fewer users overall.
- 22% of right-leaning users’ tweets contain political hashtags, compared with 12% of left-leaning users’ tweets.
- Users’ behavior is broadly distributed: many produce relatively few tweets, while a few produce substantially larger volumes.
B. Partisan Self-Identification
Right-leaning Twitter users were more likely to self-identify politically and to share hyperlinks, while their follower network was more tightly interconnected than the left-leaning network.
- Partisan Self-Identification: 38.7% of right-leaning users explicitly identified their political alignment in biographies, compared with 24.6% of left-leaning users.
- Resource Sharing: 62.5% of right-leaning political tweets contained hyperlinks, compared with 50.8% of left-leaning political tweets.Across all tweets, the corresponding figures were 43.4% and 36.5%.
- Follower Network: Right-leaning users had higher average degree, clustering, and reciprocity in same-affiliation follower subgraphs.These statistics characterize tighter interconnection among right-leaning users.
- Follower Network: The follower in-degree distributions differed significantly, while the corresponding out-degree distributions differed only marginally.Left-leaning users were roughly twice as likely to have in-degree one, whereas right-leaning users were almost four times more likely to have in-degree 1000.
- Follower Network: Right-leaning users were more densely integrated into their follower network, whereas left-leaning users were more decentralized and loosely interconnected.The left-leaning community also showed more peripheral connectivity in its k-core distribution.
B. Retweet Network
The retweet network shows structural differences favoring broader information exchange among right-leaning users, including larger retweet audiences, more sources, and more high-order network participation.
- Retweet Network: The retweet network’s node colors represented cluster assignments corresponding to politically homogeneous left- and right-leaning communities with 87% accuracy.
- Retweet Network: Right-leaning users were rebroadcast by and rebroadcast content from more individuals than left-leaning users.The statistically significant difference in degree-distribution exponents was observed at the 95% level.
- Retweet Network: Right-leaning users had a greater proportion of highly active users connected to other highly active users in high-order k-cores.The paper connects high shell index with effective information spreading under a cited SIR-based model.
- Retweet Network: A substantially higher proportion of right-leaning users participated in fully connected subgraphs of size k, known as k-cliques.The paper relates repeated exposure within such structures to political-content propagation.
C. Mention Network
Mention networks primarily captured direct political conversation, and right-leaning users showed more reciprocal mention relationships than expected by chance.
- Mention Network: 94.5% of mentions placed the target username at the beginning of the tweet, the form most strongly associated with direct conversational engagement.This supports interpreting the observed connectivity as political discourse rather than merely third-person reference.
- Mention Network: Right-leaning mention networks contained more reciprocal relationships than left-leaning networks.Relative to degree-preserving reshufflings, right-leaning reciprocal mentions occurred 7.5 times as often as expected by chance, compared with 5.6 times for left-leaning mentions.
VI. POLITICAL GEOGRAPHY
The paper estimates state-level partisan activity by comparing observed left-leaning tweet volumes with expected volumes, finding a distribution broadly aligned with traditional political geography but with notable exceptions.
- Political Geography: Fewer than 1% of Twitter users provided precise geolocation data, so the analysis relied on self-declared profile locations.The free-text field could contain arbitrary or missing entries, requiring best-guess state estimates.
- Political Geography: The geographic analysis used a cartogram whose state colors represented deviations of observed left-leaning tweet volume from chance-expected volume.Expected volume was computed from each state’s total tweets and the overall left-leaning proportion.
- Political Geography: Left-leaning users were prominent on the coasts and Northeast and underrepresented in the Midwest and Plains states.This distribution corresponded strongly to traditional U.S. political geography.
- Political Geography: Utah exhibited substantially more left-leaning content than expected by chance despite its traditionally conservative political profile.The paper presents this as an example of geographic partisan composition differing from intuitive expectations.
VII. CONCLUSION
The study identifies a shifting partisan landscape in online political engagement, with right-leaning Twitter users showing greater activity, denser social connections, and network structures facilitating information dissemination.
- Right-leaning Twitter users exhibit greater political activity, tighter social bonds, and communication networks that facilitate rapid, broad information dissemination.These findings indicate a highly active and densely interconnected right-leaning constituency.
- Right-leaning users produce more political content, devote more time to political discourse, and more often identify their political leanings publicly.
- Their community has more reciprocal relationships and broader rebroadcasting across sources.
- Right-leaning users are more likely to belong to high-order retweet network k-cores and k-cliques.
- Structural features of digital communication networks can support high-fidelity inferences about the political identities of thousands of individuals.
AUTHOR’S CONTRIBUTIONS
The authors divided responsibilities across data collection, analysis, experiment design, and manuscript preparation.
- MDC and BG collected the data and performed the analysis, while MDC, BG, AF, and FM conceived the experiments and wrote the manuscript.