Abstract.We study an agency model in which voters learn about both an incumbent and an opponent. They observe the incumbent's policy record and update their beliefs about his opponent via a campaign. Although the former is relatively more informative, it can be costly for the voter to learn about the incumbent from her policy record. This is because policy reforms, that allow a voter to learn an incumbent's ability, are risky and so can leave the voter worse off. Then the voter may prefer that the incumbent takes safer actions. The efficient level of reform-the one preferred by the voter-balances the value of learning with the expected policy cost/benefits. In a world where the opponent's campaign is uninformative reform can be too low. This is due to the incumbent's fear of failure. Or, it can be too high: the incumbent gambles on success. We show that the presence of an opponent that can reveal information via a campaign exacerbates these inefficiencies. An incumbent who anticipates the effect of an opponent's campaign on voter beliefs is more likely to make inefficient policy choices. Further, such campaigns can lead to an overall welfare loss. This is so when they do not reveal much about the opponent's ability and yet have an impact on the incumbent's policy choice.