Effective policy-making requires that voters avoid electing malfeasant politicians. However, as our simple learning model emphasizing voters’ prior beliefs and updating highlights, informing voters of incumbent malfeasance may not entail sanctioning. Specifically, electoral punishment of incumbents revealed to be malfeasant is rare where voters already believed them to be malfeasant, while information’s effect on turnout is non-linear in the magnitude of revealed malfeasance. We conducted a field experiment in Mexico, where we informed voters about malfeasant mayoral spending before municipal elections, to test whether these Bayesian predictions apply in a developing context where many voters are poorly educated and uninformed. Consistent with voter learning, the intervention increased incumbent vote share among voters with lower malfeasance priors and stronger prior beliefs, when audits revealed less malfeasance, and when audits caused voters to unfavorably update their posterior beliefs about the incumbent’s malfeasance. Highlighting the importance of information role’s in reducing the uncertainty of risk-averse voters, the incumbent party’s vote share increased even among voters that did not update their beliefs following the intervention. Furthermore, we provide evidence of heterogeneous effects of the intervention on turnout: both low and high malfeasance revelations increased turnout, while less surprising information reduced turnout. Finally, we show that party responses may also help explain our intervention’s impact.

Written with Eric Arias (William & Mary), Horacio Larreguy (Harvard University) and John Marshall (Columbia University)