A generalized logit formulation of individual choice

Abstract

Probability models of individual choice consist of two components: a formulation of random utility and the stochastic specification of that utility. Usually separable direct random utility is assumed. With Weibull error terms, logit analysis results. However, logit analysis suffers from the "assumed" "independence of irrelevant alternatives". It is the contention of this paper that these difficulties result from the usual restrictive utility formulation. A more general indirect random utility formulation is introduced. Estimates of the resulting generalized logit and the more restrictive logit models are presented. Hypothesis testing is reported which rejects the restrictive utility formulations which dominate the literature.