Source-linked AI summary
Election campaigning on social media: Politicians, audiences and the mediation of political communication on Facebook and Twitter
Sebastian Stier, Arnim Bleier, Haiko Lietz, Markus Strohmaier
TL;DR
The paper addresses the limited understanding of how politicians use different social-media platforms and whether their topics match mass-audience priorities. It combines representative survey responses with a human-interpretable Bayesian language model and social-media messages from the 2013 German election. Candidates and audiences prioritize topics differently from the mass audience, while Facebook and Twitter support different political communication purposes.
Problem
It remains unclear how politicians use different social-media platforms and whether their campaign topics address mass-audience priorities.
Method
The study trains a human-interpretable Bayesian language model on open-ended representative-survey responses and applies it to candidate and audience messages on Facebook and Twitter.
Results
Candidates and audiences discuss topics different from those salient among a mass audience, while politicians use Facebook and Twitter for different purposes.
Takeaways & Limitations
Political communication on social media is shaped by the distinct characteristics of platform audiences and sociotechnical environments.
Takeaways & Limitations
The study focuses on an election campaign in 2013, when social media campaigning was still in its infancy and micro-targeting faced severe regulatory restrictions.
Abstract
from arXiv · showhide
Although considerable research has concentrated on online campaigning, it is still unclear how politicians use different social media platforms in political communication. Focusing on the German federal election campaign 2013, this article investigates whether election candidates address the topics most important to the mass audience and to which extent their communication is shaped by the characteristics of Facebook and Twitter. Based on open-ended responses from a representative survey conducted during the election campaign, we train a human-interpretable Bayesian language model to identify political topics. Applying the model to social media messages of candidates and their direct audiences, we find that both prioritize different topics than the mass audience. The analysis also shows that politicians use Facebook and Twitter for different purposes. We relate the various findings to the mediation of political communication on social media induced by the particular characteristics of audiences and sociotechnical environments.
Introduction
The study examines whether candidates address mass-audience priorities or adapt political communication to the audiences and affordances of Facebook and Twitter. Using survey-derived topic categories and social-media messages from the 2013 German federal election, it finds different topic priorities across audiences and distinct platform uses.
- The study asks whether candidates address topics important to a mass audience and whether they tailor messages to platform-specific audiences and habits.
- The analysis compares candidates’ and audiences’ communication on Facebook and Twitter during the 2013 German federal election campaign.
- A representative survey asked respondents to describe the most important contemporary political problem in free text, producing 18 topic categories.
- A human-interpretable Bayesian language model assigns social-media messages to survey-derived topics while allowing additional social-media-specific categories.
- Candidates and audiences prioritize topics differently from the mass audience, while candidates’ topic focus is more similar to the audiences they directly encounter.
- Candidates primarily use Facebook to promote campaign activities and Twitter to comment on contemporary political events.
Related literature and research gaps
Previous research shows that online political communication is shaped by media environments and audience preferences, but cross-platform analysis of message content remains limited. This study addresses these gaps by integrating survey responses with candidate and audience messages across multiple social-media platforms.
- Research has linked media choice to perceptions of issue importance and found that online users navigate content according to personal preferences.
- Internet users pursuing political information have specific interests, while politicians have often used the web with a mass audience in mind.
- Social and algorithmic cues may intensify differences in topic priorities, and Twitter audiences during the 2013 German election diverged from survey and mass-media agendas.
- Cross-media research has often isolated one platform, focused on metadata, used small samples, or failed to categorize the topics politicians discuss.
- Existing cross-platform studies have commonly examined party accounts, attention metrics, candidate strategies, or comparisons with television rather than multiple platforms’ content.
- The study integrates multiple spheres of political communication and uses large-scale text analysis based on representative open-ended survey responses.
Strategic election campaigning on social media
The paper asks whether politicians tailor online messages to mass-audience priorities or to particular social-media audiences. It treats social media as interactive environments in which candidates’ messages may reflect their immediate communication networks.
- Politicians are expected to respond to the political preferences of their constituencies, but their online targeting remains an open question.
- Unlike campaign websites, social media place politicians in interactive contexts that may skew messages toward the preferences of their immediate communication networks.
- The study compares social-media topic salience with public opinion because national-level public agendas largely converge in the German electoral system.
- The researchers describe the study as exploratory because comparable research had not previously been undertaken.
Social media as part of multifunctional online campaigns
Social media serve multiple campaign functions beyond addressing mass-audience political concerns, including mobilization, organization, self-promotion, information, and symbolic communication. Platform audiences, affordances, and candidates’ own uses may therefore shape topic priorities and platform choice.
- Platform mediation: Social media provide campaign environments distinct from mass communication arenas, encouraging messages tailored to specific audiences despite barriers to data-driven micro-targeting.
- Campaign functions: Campaign communication can promote issue positions, demonstrate personality traits, improve name recognition, and support organizational or symbolic purposes.
- Campaign functions: Online campaigning may devote substantial attention to mobilizing supporters and organizing campaign activity.
- Audience differences: Social-media audiences differ demographically and politically from representative citizens, and German Twitter users in 2013 emphasized surveillance and campaign events over core policy issues.
- Candidate uses: Candidates’ motives range from self-promotion and information dissemination to information seeking and entertainment, reflecting varied uses of online media.
- Expectations: The paper hypothesizes that candidate and audience topic saliences on social media resemble each other more than they resemble mass-audience salience.
The use of Facebook and Twitter by election candidates
Facebook and Twitter offer different communication environments and audiences, leading candidates to use them for distinct campaign purposes. The paper expects Twitter to support policy and event commentary, while Facebook supports mobilization among interested followers.
- Platform characteristics: Twitter’s public, topic-centered architecture and retweeting enable political information to spread beyond direct follower networks.Facebook is more private, friendship-based, and algorithmically filtered, so information travels less fluidly.
- Platform characteristics: Twitter is associated with journalists, political elites, opinion leaders, and national political communication.Prior studies report greater Twitter use by national-level politicians and Swedish parties because of its potential to reach opinion leaders.
- Expected uses: Candidates are expected to use Twitter for contemporary national debates, policies, and high-attention events such as televised debates.This expectation follows Twitter’s broader public visibility and diffusion potential.
- Expected uses: Candidates are expected to use Facebook mainly for campaigning and mobilization among party supporters and local constituents.The proposed strategic value is encouraging volunteers or voter turnout rather than persuading followers about policy propositions.
- Hypotheses: The study tests whether mass-audience topic salience resembles candidates’ Twitter communication more than their Facebook communication.This comparison operationalizes the platform-specific expectation about audience reach and campaign use.
Methodology
The study combines labeled survey responses with candidate and audience messages in a Bayesian language model. This design identifies shared survey topics while allowing additional topics specific to social media communication.
- Language model: The model jointly analyzes survey responses and social media messages to classify messages into known topics and discover additional social-media-specific topics.It occupies a middle ground between supervised classification and unsupervised clustering.
- Training data: Survey answers to an open-ended question about Germany’s most important contemporary political problem provide labeled training data.The responses were coded into a hierarchical topic scheme and reorganized into 18 training topics.
- Language model: A nonparametric Dirichlet-process prior permits new Facebook- and Twitter-specific topics when survey topics cannot plausibly explain messages.Social-media topic assignments are inferred rather than observed.
- Model assumptions: The model assumes each short document is generated from a single topic, because message length limits the usefulness of mixed-membership modeling.This single-membership assumption applies to both social media messages and survey responses.
- Data limitations: Pooling 23,604 survey observations makes a larger training set but removes the sample’s initial representativeness because many participants responded repeatedly.The GLES field period ran from 8 July to 3 November 2013.
Analysis at the document level
At the document level, politicians and audiences emphasize different topics from the mass audience, while candidates use Facebook and Twitter differently. Twitter aligns more closely with public topic rankings, whereas Facebook emphasizes campaigning and mobilization.
- Platform differences: Candidates used Twitter to comment on policies and unfolding public events, while using Facebook to mobilize interested followers for campaign purposes.The results support the study’s medium-specific expectations H3a and H3b.
- Topic salience: Politicians and social media audiences mostly address different topics from those salient in the survey mass audience.This supports H1 concerning mediation between mass-audience concerns and social-media communication.
- Audience communication: Social media audiences focused overwhelmingly on Political Debates, Polity II, and Coalition Formation.These topics concern political debate, the relationship between citizens and the state, and coalition building.
- Politician–audience alignment: Politicians and audiences were more synchronized on Twitter, whereas Facebook showed a considerable disconnect between candidates’ campaigning and audience discussion.Although politicians devoted 42.3% of Facebook messages to campaigning, their audiences mostly discussed other topics.
- Mediation: Topic ranks across social-media corpora correlated more strongly with one another than with the survey, indicating substantial mediation by engaged audiences and platform transmission mechanisms.Politicians adjusted their communication to the mediated environments to which they were most directly exposed.
- Public agenda: Despite differences in topic shares, politicians and audiences discussing policies on social media tended to prioritize similar topics to representative survey respondents.The authors therefore describe the public agenda as relatively cohesive across media during the campaign.
Analysis at the word level
The word-level analysis compares topic-word distributions across corpora rather than only comparing topic shares. It finds that survey language differs from social-media language, while platform and actor contexts also shape how topics are discussed.
- Contribution: The word-level analysis makes topic generation more transparent and increases understanding of mediation processes in political communication.It moves beyond topic salience toward comparing perspectives on political topics across audiences and media.
- Word-level comparison: Cosine similarities between corpora vary considerably by topic, showing that topic overlap does not guarantee similar ways of discussing that topic.The analysis uses topic-word distributions on a 0-to-1 similarity scale.
- Survey versus social media: Corpus pairs including the survey had approximately 12% lower cosine similarity, indicating markedly different descriptions of political problems.The comparison controls for corpus-pair token counts and other structural factors in regression models.
- Platform effects: Politicians’ tweets were significantly more similar to other corpora than the average corpus pair, unlike politicians’ Facebook posts.Twitter candidates emphasized survey-relevant aspects while also using language similar to other social-media content layers.
- Limitations: The interpretation of Twitter communication may be limited because tweets can reflect platform conventions and space restrictions rather than strategic considerations.The authors test this alternative explanation with audience and politician indicators in additional models.
- Robustness: The results remained after controlling for whether the actors were the same, supporting distinct communicative practices across media.The Same actor variable was insignificant, while platform effects persisted.
- Platform effects: Same-medium corpus pairs had significantly higher similarity, with p ≤0.009 in all models.This indicates that politicians and audiences addressed similar aspects when discussing topics on the same platform.
- Topic types: Political-process topics were more semantically diverse than polity or policy topics.Similarities did not differ systematically between survey-known and newly created social-media topics.
Conclusion
The study shows that politicians and audiences discuss different topics on social media than the mass audience, while Facebook and Twitter support distinct communication purposes. These findings indicate that political communication is strongly mediated by platform-specific sociotechnical environments, limiting broad generalizations from one platform.
- Conclusion: The study compares social-media topics with those most important to a mass audience using survey and social-media text analysis.The analysis uses a model developed to apply survey-derived topic categories to social-media messages while allowing additional social-media-specific topics.
- Conclusion: Politicians and audiences discuss campaign-related events and social-media-specific topics more often than policies, and their salient topics resemble each other more than survey respondents’ priorities.When policies are discussed, candidates and audiences prioritize them similarly to representative-survey respondents, suggesting an integrated public agenda in that domain.
- Conclusion: Candidates use Twitter for quasi-public masspersonal communication and Facebook for more direct organizational and mobilization purposes.The study relates these different purposes to the distinct target groups candidates encounter on each platform.
- Conclusion: The study’s scope is limited by its focus on one 2013 German election campaign, and it remains uncertain whether the findings apply to other countries or the near future.The authors also note that demographic information on politicians’ follower graphs was unavailable and that multidimensional messages had to be assigned to one topic.
- Conclusion: Social-media platforms and their sociotechnical environments strongly mediate political communication, producing relevant differences between platforms.The findings suggest that campaign strategies and political communication are shaped by varying platform affordances.
Notes
The appendix documents the survey inputs, topic-model specification, inference procedure, and robustness checks underlying the social-media analysis.
- Survey inputs: Survey participants reported the most important political problem before and after the election, with response rates documented for both waves.Before the election, 7,249 participants stated a first problem and 6,673 a second; after it, 5,016 and 4,666 did so.
- Robustness: Robustness checks largely confirm the main findings across alternative topic sets and model specifications.The exception is that Twitter politician and audience dummy variables lose significance in the model using only known topics.
- Model: Each document is generated by drawing a topic from global topic popularity and then words from the selected topic’s vocabulary distribution.The model uses topic assignments and multinomially distributed words, with Dirichlet priors α and β.
- Model: The model jointly represents GLES responses and social-media messages, with shared topic-generation assumptions but observed labels only for GLES documents.Social-media topic assignments remain unobserved and can extend beyond the initial GLES topics.
- Inference: Inference uses collapsed Gibbs sampling to iteratively sample social-media topic indicators after integrating out other variables.The procedure assumes convergence after sufficient iterations, with the resulting topic distribution independent of initialization.
Appendix C: Survey dataset
Appendix C constructs a training dataset from GLES responses by refining survey topic categories into 18 discriminative topics.
- Topic construction: The training topics are designed to be sufficiently discriminative and relatively equal in size.The resulting topic coding is documented as reproducible from Table A1.
- Survey dataset: GLES responses were coded into politics, polity, or policy areas, with three small categories removed from the analysis.Removed categories covered other problems, East Germany, and cultural leisure policy.
- Topic construction: The resulting topic areas ranged from 188 observations for politics to 6,010 for social policy.Some areas were split into subtopics after testing combinations and qualitatively assessing topic overlap.
- Topic construction: Social policy was divided into general social policy, family policy, health care and pensions, and migration and integration.Fiscal policy was further divided into general fiscal policy, currency and euro, taxes, and budget and debt.
- Topic construction: Foreign and defense policy were merged into Foreign Policy (Defense), while Europe was retained as a separate topic.The final training set contained 18 discriminative topics covering 23,295 responses.