Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
Abstract I show that persistent underconfidence and overconfidence can each arise from rational Bayesian learning when effort and ability are complementary. Which arises depends on the decision-making environment, and in particular on the effect that greater effort has on the variance of outcomes. Agents learn away overconfidence and underconfidence at asymmetric rates because (i) Bayesian updating requires that their sensitivity to new information depend on their effort choices and (ii) their effort choices in turn depend on beliefs about their own ability. As one implication, I show that management can credibly induce additional effort from employees by designing feedback that generates average overconfidence through being conditionally vague.