Discrete choice under risk and model uncertainty

This paper studies the aggregation of discrete choice models under risk. We propose a unanimity condition: if every model predicts one item is chosen more frequently from a menu than another, so should the aggregate model. We show that this condition entails the aggregate model must be a mixture of the individual models. This result provides a behavioral characterization of the Bayesian model averaging method for discrete choice models under risk. As an application of the result, we examine the aggregation of a continuum of probability measures.