Current Issues in Discrete Choice Modeling

Until recently, computational constraints forced researchers in the discrete choice area to limit themselves to very simple statistical models, such as the multinomial logit (MNL), in which choice probabilities could be evaluated quickly on a computer. But the MNL only makes sense as a behavioral model under very special circumstances. Recent advances in computation make it possible to estimate richer behavioral models that generate very complex choice probability expressions. This paper discusses a number of possible avenues for future research in the discrete choice area in light of these developments