Estimating the Impact of Political Propaganda on Reddit Users’ Political Opinions

Abstract

This paper applies supervised aggregated sentiment analysis and fixed effects models to comments on the popular social media platform Reddit to assess the impact of digital propaganda on users’ sentiments toward politicians ahead of the 2016 U.S. election. While some studies have attempted to quantify the impact of digital propaganda, none have been successful so far. Reddit also remains an under studied source of information about political opinions online and especially the impact of propaganda. For this paper, I used a set of accounts Reddit had identified as having engaged in coordinated inauthentic behavior starting before the 2016 election that are believed to originate with Russia’s Internet Research Agency, the body behind a significant portion of digital propaganda aimed at interfering with U.S. politics. Users who responded to posts made by these accounts formed my treatment group, and users who did not but who had used Reddit in similar ways formed my synthetic control group. A fixed effects model varying over time at the quarter level helped to further limit the impact of any time-invariant systematic differences between groups. As was commonly believed, it appears that exposure to propaganda predicted greater favorable comments about Donald Trump and fewer favorable comments about Hillary Clinton. However, contrary to expectations, I found no evidence of a shift in favorability toward Sen. Bernie Sanders, indicating that the IRA focused primarily on the two main candidates rather than using the senator as a wedge issue to divide potential democratic voters.