We investigate how the gender gap in confidence affects the views that evaluators (e.g., employers) hold about men and women. We find the confidence gap is contagious, causing evaluators to form overly pessimistic beliefs about women. This result arises even though the confidence gap is expected and even though the confidence gap shouldn’t be contagious if evaluators are Bayesian. Only an intervention that facilitates Bayesian updating proves (somewhat) effective. Additional results highlight how similar findings follow even when there is no room for discriminatory motives or differences in priors because evaluators are asked about arbitrary, rather than gender-specific, groups.

Myriad environments allow for the possibility of confusion. Agents may appeal to such confusion—or the possibility of making an honest mistake—to justify their behavior. In three sets of experiments involving thousands of subjects, we document evidence of such motivated errors. We document this evidence in a simple environment in which the scope for errors is small and in more complex environments in which subjects display correlation neglect and an anchoring bias.