Polarization, Partisanship and Junk News Consumption over Social Media in the US

Vidya Narayanan; Vlad Barash; John Kelly; Bence Kollanyi; Lisa-Maria Neudert; Philip N. Howard

In brief

This study maps the distribution of "junk news" (extremist, sensationalist, conspiratorial, and professionally non-credible outlets) across Twitter and Facebook networks before a major US political event. Using network clustering and heterophily analysis, the authors find that Trump supporters on Twitter share the widest range of junk news sources and disproportionately circulate such content; right-wing groups show similar patterns on Facebook. The findings reveal stark partisan asymmetry in misinformation consumption.

Contributions

Developed a grounded typology of junk news sources (91 outlets) across five dimensions: professionalism (lack of editorial standards), style (emotionally driven language), credibility (reliance on conspiracy theories), bias (hyper-partisan reporting), and counterfeiting (mimicry of legitimate news media).
Applied social network mapping with Fruchterman-Reingold force-directed visualization to identify and visualize political audience clusters on both Twitter and Facebook.
Demonstrated quantifiable partisan asymmetry: Trump supporters on Twitter have 95% coverage of known junk news sources and account for 55% of junk news traffic in the network.
Computed heterophily scores as a measure of cross-group connection strength, revealing stark partisan clustering and differential engagement with misinformation across the political spectrum.
Analyzed platform differences: Twitter audiences engage with a wider range of junk news sources than Facebook audiences, and partisan patterns differ between platforms.

Method

Data collection and sample:

Collected 22,117,221 tweets during the US election period (November 1–11, 2016) using a curated seed list of propaganda websites across the political spectrum.
Identified 13,477 Twitter user accounts through network association and reduced this to a "discussion core" of 12,413 highly connected users via k-core reduction (k=1).
For Facebook, harvested public page data from 47,719 public pages and reduced to a well-connected sample of 10,691 pages.

Network analysis:

Used hierarchical agglomerative clustering with bipartite graph structural similarity metrics to segment users/pages into distinct communities.
Clustered Twitter users by following/follower relationships; Facebook pages by "like" relationships.
Computed heterophily index for each group pairing: a ratio-of-ratios that measures whether two groups have more or fewer inter-group connections than expected by chance. Natural log transformation balances the index across the range.

Junk news classification:

Team of 12 trained coders systematically labeled 91 sources as junk news if they met at least three of five criteria: lack of professional journalism standards, emotionally driven / hyperbolic language, reliance on false information and conspiracy theories, biased / hyper-partisan reporting, or counterfeiting of legitimate news media.
Assigned coverage and consistency scores: coverage = percentage of all junk news domains a group linked to; consistency = percentage of junk news links from a group shared by that group.

Results

Twitter findings:

Trump Support Group: 95% coverage of junk news sources, 55% consistency (accounts for majority of junk traffic).
Conservative Media Group: 95% coverage, dominant role in spreading junk news.
Democratic Party Group: 40% coverage, mainly engages with mainstream outlets.
Heterophily scores reveal strong clustering: Trump Supporters and Conservative Media (1.4), Republican Party and Trump Supporters (1.4), Democratic Party and Mainstream Media (1.7).

Facebook findings:

Hard Conservatives Group: 91% coverage, 58% consistency (disproportionate junk news traffic).
Military/Guns Group: 45% coverage.
Conspiracy Group: highly connected to fringe communities (heterophily ~2.5).
Democratic Party and Mainstream Media show 40% coverage and substantial engagement with legitimate sources.

Cross-platform comparison:

Twitter audiences engage with a wider range (greater coverage) of junk news sources than Facebook audiences.
Partisan polarization more pronounced on Twitter; Facebook shows more mixed engagement patterns.
Right-wing accounts and groups more densely interconnected with multiple junk news outlets across both platforms.

Notes

A landmark COMPROP data memo that provides precise measurement of junk news consumption asymmetries across the two dominant US social platforms. The heterophily index is a novel metric for understanding partisan clustering. The study's reliance on a curated seed list of junk news sources (rather than algorithmic detection) means coverage is limited to known outlets—emerging or novel misinformation sources would not be captured. The 2016 election context is specific, but the methodological approach remains influential for studying polarization and misinformation networks.