Model selection and misspecification in discrete choice welfare analysis

This study extends the work by Herriges and Kling (1997) to further evaluate the impact of discrete choice modelling techniques on welfare measures. Particularly, we evaluate the performance of the increasingly popular mixed logit model and the computational strategy for deriving discrete choice welfare measures. Our simulation results show that model misspecification can have profound effects on welfare measures. In general, the flexible mixed logit model performs relatively well in the presence of misspecification. However, when the nesting structure can be appropriately identified (via statistical tests and a priori knowledge/experience), the nested logit model provides more reliable welfare measures than the mixed logit model.