Is the distribution of resolvable uncertainty Type I extreme value? A Test for Random Coefficient Models using Choice Probabilities

Abstract

Stated choice probabilities are increasingly used in conjunction with the random-coefficient model (RCM) to describe individual preferences. They allow survey respondents to express uncertainty about the future or the incompleteness of a hypothetical scenario: the resolvable uncertainty. Parametric assumptions such as a Type I extreme value (EV1) distribution are almost always imposed on this uncertainty to identify and estimate the associated RCM. This paper proposes the first test for these parametric assumptions, based on a nonparametric identification result for the population distribution of the interquantile range of the resolvable uncertainty. In all four empirical applications considered, the test finds strong evidence against the EV1 assumption.