Polarization and echo chambers in Reddit's political discourse

Abstract

Political debate nowadays takes place mainly on online social media, with election periods amplifying ideological engagement. Reddit is generally considered more resistant to polarization and echo chamber effects than platforms like Twitter or Facebook. Here, we challenge this assumption through a case study across the 2016 US presidential election. We use statistical validation techniques to extract ideologically distinct communities of subreddits, in terms of their contributing user base and news consumption, which we use to analyze the dynamics of political debate. We thus reveal clear polarization in both interaction-based and topic-based communities, with clusters of Democratic, Conservative, and Banned subreddits. Election periods intensify cross-group engagement, align Banned and Conservative content, and reduce linguistic diversity within groups. Overall we characterize Reddit as a polarized environment marked by the presence of echo chambers, highlighting network validation as a key method for identifying behavioral and interaction patterns on online social media.

Figures (62)

• Figure 1: Methodological framework. This figure summarizes the main approaches used in this study, analyzing two dimensions--news domains and user interactions--to reveal the presence of echo chambers. a Structure of Reddit data: conversation trees with root posts and comments underneath, belonging to different politics-related subreddits. b Construction of the "user-interaction" network: from the bipartite network of users interacting on subreddits to the validated network of significant similarities among subreddits (in terms of common user base). c "Information-diet" approach: from the bipartite network of news domain shared by subreddits to the validated network of significant similarities among subreddits (in terms of commonly shared domains). d Matching between the communities of subreddits in the two validated networks reveals echo chambers: closed discussion forums characterized by overlapping user and news source patterns.
• Figure 2: Temporal evolution of subreddits' communities with similar user base. a) Sankey diagram where the width of the flows is proportional to the number of subreddits moving among communities. Community colors reflect the composite hues of their underlying subreddit tags, shown in the legend. Label "Not Validated" refers to subreddits that are not connected to others by statistically validated links. b) Evolution of the main communities of the validated subreddits, with several communities exhibiting strong tag homogeneity. c) Focus on Democratic, Conservative, and Banned subreddits. As elections approach, both Democratic and Conservative subreddits increase in number, while Banned subreddits exhibit more frequent co-occurrences with Conservative ones.
• Figure 3: Polarization, engagement and linguistic similarity across subreddit communities. a Donuts charts showing the distribution of user labels within the communities of the validated "user-interaction" network projections, for 2014 and 2016. b Donuts charts for tag-based communities, for 2014 and 2016. In both cases, polarization decreases in time, with some exceptions (e.g., Democratics and Banned). c Bar plots showing annual polarization levels across a selection of communities, with solid bars computed for all users and dashed bars computed excluding "Banned" users. This distinction allows to capture the strong polarizing role of banned subreddits, especially within conservative clusters. Early polarization is most pronounced among Far-Right and Democratic groups, with major shifts observed in the Conservative category. Over time, overall polarization decreases, except for a notable rise among banned communities. d Average scores of comments for 2014 and 2016, revealing a general decline in user engagement, particularly in Democratic, Conservative, and Banned subreddits. e Cosine similarity between posts within topic-based subreddit groups, indicating strong intra-group similarity and increasing inter-group similarity over time, especially between Democratic and Conservative communities.
• Figure 4: Patterns of News Consumption and Echo Chambers. a: Sankey diagram where the width of the flows is proportional to the number of subreddits moving among communities, defined according to similarity of shared news sources among subreddits. Some communities remain highly homogeneous in terms of tags, while a merging of themes occurs in certain years. For example, Democrats and Conservatives initially coexist in the same community, but beginning with the election year, they gradually separate into distinct groups, eventually forming separate Democrats and Conservatives/Banned communities by 2017. b: Chord diagrams depicting the overlap (in terms of common subreddits) between communities identified through the "user-interaction" and "information-diet" analyses, for 2014 and 2016 respectively. The color of the flows represents the average of the colors of the source and destination communities. The coherent structure of the communities defined by the two approaches is indicative of the presence of echo chambers in political debate. Indeed the strongest overlaps tend to occur between communities with similar political or thematic orientations. Additionally, the figure shows the validated networks derived from the chord diagrams.
• Figure 5: Engagement and Temporal Dynamics among Democratic, Conservative, and Banned Subreddits. a: Yearly textual distances between Democratic, Conservative, and Banned subreddits. We report yearly text-embedding cosine distances between Democrats and Conservatives, which decrease in 2016, particularly among politically active users. Both groups also converge linguistically toward Banned subreddits over time, although Democrats remain comparatively more distant. As benchmarks, we include a random model (the mean similarity of randomly assembled communities of the same size) and a heterogeneous model (the mean of all pairwise similarities among a community’s constituent subreddits, thereby controlling for internal composition heterogeneity). b–c: Distances in user- and domain-based subreddit networks. We show average pairwise distances (normalized by yearly network averages) in statistically validated interaction networks, computed using the harmonic mean. A widening gap emerges between Democrats and Conservatives, with Banned subreddits consistently closer to Conservatives. Domain-sharing networks reveal a similar pattern: Democrats and Conservatives diverge sharply after 2016, while Banned subreddits remain aligned with Conservatives. d–e: Heatmaps of yearly proportions of comments and scores exchanged within and between user groups (shown for 2014 and 2016). They indicate increasing cross-faction interaction over time, but consistently lower comment scores for inter-group exchanges compared to intra-group ones.