Proceedings of the Annual Meeting of the Cognitive Science Society

Bayesian models that optimally integrate prior probabilities with observations have successfully explained many aspects ofhuman cognition. Research on decision-making under risk, however, is usually done through laboratory tasks that attemptto remove the effect of prior knowledge on choice. To test the effects of manipulating prior probabilities on participants’choices, we ran a large online experiment in which risky options paid out according to the distribution of Democratic andRepublican voters in unknown congressional districts in known US states. This setup allows us to directly manipulate priorprobabilities while holding observations constant and to compare people’s choices with the options’ true posterior values.We find that people’s choices are appropriately influenced by prior probabilities, and discuss how the study of risky choicecan be integrated into the Bayesian approach to studying cognition.