Estimation of discrete choice models with error in variables: An application to revealed preference data with aggregate service level variables

Abstract

This article proposes a method to estimate disaggregated discrete choice models with errors in the variables. The objective is to estimate the discrete choice models' coefficients to compute the value of time and use it for cost-benefit analysis in transportation planning. The method is general, as it only requires a validation sample to estimate the conditional density of the error-free variables given the mismeasured variables. More specifically, we assume that the attributes of the chosen alternative are reported without error in revealed preference surveys, and we use this information as the validation sample. The mismeasured variables may be spatially aggregate service levels from mobility surveys or transportation network models. Monte Carlo simulations show that the proposed method substantially reduces bias in parameters. We validate the technique with two real data sets from Santiago, Chile.