Proceedings of the Annual Meeting of the Cognitive Science Society

Stereotypes are pervasive and rigid. We propose that stereotyping arises because perceivers solve a hierarchical inference problem in which traits are inferred simultaneously at the individual and group levels. Using hierarchical Bayesian models, we derive normative predictions about stereotype formation and updating and test them in three experiments. Study 1 shows that evidence pooling in hierarchical inference enables group-level impressions to form faster than individual-level impressions when perceivers observe limited information of many group members, allowing strong stereotypes without strong individual impressions. Study 2 shows that stereotypes resist updating when stereotype-inconsistent group members behave heterogeneously, making perceivers attribute counterevidence to individual differences rather than updating group stereotypes. Study 3 shows that when stereotype-inconsistent individuals also belong to a second, unfamiliar group, counterevidence is attributed to this second group, leaving the original stereotype intact. Together, these findings provide a computational account of how hierarchical inference produces and sustains rigid group beliefs.