Learning Gaps? Attitudes and Beliefs Under Described and Experienced Signals

Abstract

We show that ambiguity attitudes vary systematically with the informativeness of available signals, using an experimental design that varies informativeness and a method that separately measures attitudes and beliefs. When signal-event associations are uncertain and must be inferred from sampling (experience), aversion decreases as signals become more informative, and insensitivity drops sharply under the strongest signal. By contrast, when associations are fully known (description), aversion is essentially unchanged across informativeness levels, while insensitivity shows at most a modest response. This contrast suggests the informativeness effect reflects responses to ambiguity about signal-event associations. Separately measuring attitudes and beliefs proves consequential: the two can move in different directions, and failure to separate them overstates conservatism in updating. For beliefs, subjects in description underweight signal information relative to the Bayesian benchmark, while subjects in experience cannot be distinguished from it under our baseline prior specification, though this finding is sensitive to the assumed prior.

Keywords: Ambiguity Attitude, Signal Informativeness, Decision from Experience, Description-Experience Gap, Belief Updating, Learning C91, D81, D83