Estimating Nested Logit Models with Censored Data

Abstract

The presented methodology estimates parameters of a nested logit model when decision makers who choose one of the alternatives are systematically excluded from the sample data that are used to estimate the model (i.e., censored). Unlike existing methods for estimating discrete choice model parameters with censored data, which require exogenous information beyond the specification of the model to be estimated and the available sampled observations, the proposed method requires no additional outside information. It is demonstrated empirically that this approach can recover not only generic model parameters that apply to common attributes of all alternatives but also parameters for alternative-specific constants and variables associated with both observed and censored alternatives. Although the standard errors of the estimated parameters are larger than those of models estimated with uncensored data, censored data methods still hold great potential for applications in which uncensored data are expensive or impossible to collect.