Association for Public Policy Analysis & Management

This paper presents the first analysis of the relationship between electoral crimes and electoral performance in large democracies. Using a sample of candidates charged with electoral crimes in the race to municipal office in Brazil between 2004 and 2016 and employing an instrumental variables strategy, I find that being ultimately convicted by the Brazilian Electoral Court reduces the probability of election by 28.8 percentage points and convicted candidates’ vote share by 17.9 percentage points. This result suggests that judicial decisions are strong information signals about a candidate’s type and voters punish convicted candidates, in line with other studies showing that information quality matters for electoral accountability. In addition, this paper implements machine learning techniques to code judicial sentences and to provide heterogeneous treatment effects of various electoral crimes, a novel approach in the comparative judicial politics literature.