The use of choice models is widespread in transport studies. A review of existing choice models includes a critical description of multinominal logit, structured logit, and multinominal probit models. It is concluded that although the logit models are adequate for some purposes, they are unable to provide a good enough representation of reality when correlation between alternatives is significant, as in some aspects of modal split and route choice modelling. In these cases the multinominal probit model, based on multivariate normal cost distributions, appears to provide an acceptable alternative. Unfortunately analytic solutions are not available for problems with more than two choice alternatives, and recourse has to be made to approximate solutions, with their attendant problems of long computation time or inaccuracy. (Author/publisher)
Abstract