Choice model specification, substitution and spatial structure effects

Abstract

Several choice models are compared on their ability to reproduce two types of simulated data sets. The sets belonging to the first type were generated by using a probit model which is able to account for substitution effects while the data sets of the second type were generated by using a probit model which is able to account for spatial structure effects. The main conclusion of the experiment is that simple models like the multinomial logit model, although they perform less than the models used to generate the data, are sufficiently robust to reproduce the simulated data.