Four types of typical discrete Choice Models: Which are you using?

Abstract

Four types of typical discrete Choice Models: Multinomial logit (MNL) model, Nested logit (NL)model, Heteroscedastic Extreme Value (HEV)Model and Mixed logit model, have been proposed and implemented in empirical investigations, although there is no universally acknowledged using principle. Here we report study to test this type of models in a travel mode choice case. We implemented four models calibration using software programmed by ourselves. We found that if sample data satisfied with IIA property, our experience has confirmed that MNL is the first choice in mode split forecasting. The nested logit model and Heteroscedastic Extreme Value (HEV)Model are not significantly better than the multinomial logit model. Mixed logit model corrects IIA flaw, but is somewhat more difficult to estimate.