p.1289

Maximum Likelihood Estimator for Choice-Based Samples

Stephen R. Cosslett

Abstract

A discrete-choice probability model can be estimated from a sample stratified on the choice variable by maximizing the "pseudo-likelihood," a quantity closely related to the log likelihood for a random sample. We investigate the asymptotic properties of the estimator, and show that it is consistent, asymptotically normally distributed, and satisfies a commonly used criterion for asymptotic efficiency.