Partly as a result of the increasing reliance on Stated Choice (SC) data,the vast majority of discrete choice modelling applications are now estimated on data containing multiple observations for each respondent. At the same time there has been growing interest in the representation of unexplained heterogeneity in choice data, using random coefficients models such as Mixed Multinomial Logit (MMNL). The presence of multiple observations for each respondent can indeed be a great asset in the identification of such variations in tastes. However, in this paper, we question the validity of the common assumption that tastes vary across respondents but stay constant across repeated choices for the same respondent. We extend the existing framework for the MMNL analysis of panel data by allowing for intra-respondent heterogeneity on top of inter-respondent heterogeneity. An empirical analysis making use of a SC dataset for route choice confirms our hypotheses and shows that superior performance is obtained by our more general model. (A) Reprinted with permission from Elsevier.
Abstract