Citizens are expected to punish corrupt politicians at the polls. In line with this presumption, lab and survey experiments consistently show that citizens are unlikely to vote for candidates that engage in corruption. At the same time, observational studies and field experiments frequently conclude that corrupt politicians are only mildly punished by voters. This apparent contradiction could be due to the design implemented in previous lab and survey experiments. In the real world, individuals tend to avoid and downplay information that challenges previously held beliefs, like their partisanship. An experimental design that randomly informs participants about corruption, and disregards the fact that citizens are prone to self-select information, is highly unrealistic and might lead to an overestimation of the electoral consequences of corruption. Unlike previous studies, this paper implements a Preference Incorporating Choice and Assignment (PICA) experimental design in order to address the following question: how does information about corruption affect the likelihood of voting for corrupt politicians when accounting for information self-selection? Based on an online experiment conducted in Spain (N = 3,200), the PICA design allows us to estimate the electoral consequences of corruption accounting for the fact that citizens are able to overlook information about corruption from their preferred party, either by exposing themselves selectively to information about the good performance of their party or by avoiding information about politics altogether. The PICA design increases validity by explicitly modelling how citizens navigate information about malfeasance from their party, while retaining the internal validity of fully randomized experiments.