Elusive Beliefs: Why Uncertainty Leads to Stochastic Choice and Errors

People often cannot assign a clear probability to an event but face uncertainty about their subjective probabilities. We model belief uncertainty by assuming that agents’ beliefs are characterized by a distribution over subjective-probability distributions that agents cannot access directly. Our model produces stochastic choice because each decision-relevant belief is but one realization out of the distribution over all possible beliefs. Our model predicts that when comparing unknown situations to routine choices, people will make more ex-ante suboptimal choices in unknown situations. The model also offers an explanation for experiment participants not playing a best-response to their stated beliefs: participants are uncertain what belief to report or base their decision on, and hence, act on momentaneous ‘belief realizations’. In an experiment, we exogenously manipulate participants’ belief uncertainty. We find support for both predictions. Low belief uncertainty leads to fewer errors and thus, higher earnings, even when controlling for the accuracy of participants’ beliefs. Second, under low belief uncertainty, observed best response rates are high and increasing in the amount of information we provide. Conversely, high belief uncertainty leads to lower consistency.