Why do political leaders or managers persist with their pet projects and policies despite bad news? When project continuation is a more informative experiment than project  termination, a reputationally concerned leader is biased towards continuation, as it enables her to disclose her private information. Perceived overconfidence on the part of the leader aggravates this tendency, even when the leader is not, in fact, overconfident. Higher-order beliefs regarding overconfidence can induce inefficient equilibrium selection  even when it is "almost common knowledge" that the leader is not overconfident. Thus, a culture where leaders are expected to be overconfident can have undesirable effects even upon leaders who have correct beliefs.