The Heterogenous Logit Model

Probabilistic choice systems in the generalized extreme value (GEV) family embody two restrictions not shared by the covariance probit model. First, the unobserved components of random utility are homoscedastic across individuals and alternatives. Second, the degree of similarity among alternatives is also assumed to be constant across individuals. This paper considers extensions to models in the GEV class which relax these two restrictions. An empirical application concerning the demand for cameras is developed to demonstrate the potential significance of the heterogenous logit model.