Retrospective voting circumvents many of voters' cognitive limitations, but if voters' attributional judgements are systematically biased, retrospective voting becomes an independent source of political failure. We design and administer a new survey of the general public and political experts to test for such biases. Our analysis reveals frequent, large, robust biases, with an overarching tendency for the public to overestimate politicians' ability to influence outcomes. Retrospective voting usually gives elected leaders supraoptimal incentives, though there are important cases where the reverse holds.