Proceedings of the Annual Meeting of the Cognitive Science Society

People routinely infer others' competence under uncertainty, often relying on cues such as task difficulty and past accuracy. An emerging body of research suggests that people approximate Bayesian inference when doing so. We extend these results by testing whether people can infer others' numerical ability in a way that is consistent with a rational Bayesian model. In Study 1, we find that participants accurately predict the arithmetic performance of another individual from information about their past performance. Computational modeling shows that participants' inferences are better described by Bayesian processes than by plausible heuristics. Study 2 introduces a modified paradigm, in which participants are told about both past performance and time taken to solve problems. We find that, although participants are quite accurate in their predictions, they do not seem to take into account information about speed.