The paper considers a repeated election game between an infinitely-lived representative voter and finitely-lived, heterogeneous politicians. The voter's prior belief about the incumbent's competency is updated during the incumbent's first term in office. The voter's problem is to find a rule that simultaneously selects and controls politicians. We show that the simple performance rule, standard in the literature, is justified as a time-consistent rule for a forward-looking voter. The outcome of a large class of perfect equilibria is “strategic caution”: incumbent politicians slow down the voter's Bayesian learning by taking only weakly informative actions.