Some lessons for working with repeated choice data

Abstract

The vast majority of discrete choice modelling (DCM) applications are now estimated on Stated Preference (SP) data, including, but limited to in the field of transport research. In SP data, each respondent is faced with multiple choice situations, and recognising this repeated choice nature of the data is a crucial modelling issue. With the increasing popularity of the Mixed Multinomial Logit (MMNL) model, most applications now rely exclusively on a random coefficients approach in dealing with the repeated choice nature of the data. Here in turn, the assumption is generally made that tastes vary across respondents, but not across observations for the same respondent. In this paper, we question this assumption and show that it is important to also allow for variation in tastes across replications for the same respondent. Furthermore, the paper shows the benefits of accounting for correlation across replications for the same individual independently of the treatment of taste heterogeneity.