Abstract
<jats:p>We study how politicians' adoption of social media, and the use of moral language, affect electoral returns. We compile a new dataset covering over 4,000 Twitter accounts and almost five million tweets by US House candidates over 2008-2018. We document several descriptive facts: social media use rises and falls with electoral cycles, politicians' use of moral rhetoric increased over the period of study, and moral rhetoric is strongly associated with higher social media engagement. Using a regression discontinuity design on close primary elections, we find that social media use by candidates has a positive but imprecisely estimated effect on vote shares. This effect is strongly heterogeneous depending on the use of moral rhetoric, and is large and statistically significant for candidates with above-median moral language use. A complementary design using only candidates with Twitter also reveals a causal effect of moral language on vote shares. We also document that relative to other social media features (negativity, out-group animosity, toxicity) moral rhetoric drives more positive engagement. Finally, we show that increased coverage from mainstream news media and campaign donations are important mechanisms.</jats:p>