That citizens engage in retrospective voting is widely established in the literature. But to what extent is retrospection affected by the expectations that leaders set in advance? We develop a theoretical framework of how expectation setting affects voters' retrospective evaluations of incumbent performance. To test the theory, we conduct a series of between-subjects experiments in which we independently manipulate both expectation setting and the eventual outcome. In domains where politicians have practical authority, or direct influence over outcomes, setting high expectations incurs a cost in public support if the projected outcome is not attained. The same is true in domains where politicians have theoretical authority, or limited influence, but where expectation setting sends a signal about the leader's judgment. However, in domains where politicians have neither practical nor theoretical authority, setting high expectations is unambiguously beneficial, implying that optimism is valued by voters as a personality disposition.