Partial identification of counterfactual choice probabilities

This article shows how to predict counterfactual discrete choice behavior when the presumed behavioral model partially identifies choice probabilities. The simple, general approach uses observable choice probabilities to partially infer the distribution of types in the population and then applies the results to predict behavior in unrealized choice settings. Two illustrative applications are given. One assumes only that persons have strict preferences. The other assumes strict preferences and utility functions that are linear in attribute bundles, with no restrictions on the shape of the distribution of preference parameters.