Seminar Political debate on social media: theory and evidence

Abstract

We combine theoretical and empirical analysis to study online political debate under low trust, widespread toxicity and disinformation. Our model explores how users strategically employ aggressive language ("hate speech") and corroborating evidence. Using a semi-supervised learning approach, we estimate users’ ideological distance in about 140,000 Twitter interactions. Aggressive language in replies increases with ideological distance, while evidence use (hyperlinks) follows an inverted U-shape; they co-occur in ways consistent with our model predictions. We argue that analysis and policy discussion should focus on the context-dependence of behavior rather than only users’ individual traits.