How do voters' behavioural biases affect political outcomes? We study this question in a model of Downsian electoral competition in which candidates have private information about the benefits of policies, and voters may infer candidates' information from their electoral platforms. If voters are Bayesian, candidates ‘anti-pander’ – they choose platforms that are more extreme than is justified by their private beliefs. However, anti-pandering is ameliorated if voters' inferences are subject to confirmation bias. Voter confirmation bias causes elections to aggregate candidates' information better, and all observers, whether biased or Bayesian, would like the voters in our model to exhibit more confirmation bias than they do themselves.