This article addresses simultaneously two important features in random utility maximisation (RUM) choice modelling: choice set generation and unobserved taste heterogeneity. It is proposed to develop and to compare definitions and properties of econometric specifications that are based on mixed panel Logit (MXL) and latent class panel Logit (LCL) RUM models in the additional presence of prior compensatory screening decision rules. It is also proposed to evaluate and to test each against the other ones in an application on home-to-work mode choice in the French Parisian region using 2002 data.
Choice set formation is defined as a behavioural process that models the way some alternatives may be screened out before a final decision is made between the remaining ones. Actually, given a full description of available choice alternatives, one does not observe whether all of them are considered or not for final decision. One may assume that only a non empty subset of them is selected before making it. Even though each available alternative could be chosen, some may consciously be screened out to facilitate the decision task. A quick look at the literature shows that it has been proposed several strategies to model prior selection of choice alternatives within a RUM framework, e.g. conjunctive and disjunctive rules, penalized utility function, multivariate latent bounds, parametric independent availability model (PIAM) or parametric consideration set (PCS), and single latent threshold (SLT). PIAM/PCS and SLT are the two screening rules that will be used and compared in the paper. PIAM and PCS assume that an alternative is considered in the final choice set if and only if its associated level of utility is larger than a specific latent level of consideration for it. SLT considers that all the alternatives in the final choice set give larger levels of utility than a unique latent level of consideration. Whichever is the screening rule, it is assumed that the alternatives that pass the screen are evaluated in a manner consistent with random utility theory (RUM choice models). It is also assumed that the considered set of choices cannot be empty.
Econometric specifications are defined as LCL and MXL models with endogenous choice set generation processes that are based on PIAM/PCS and SLT screening rules. Applications of the PIAM and PCS approaches have already been made using MXL models. One does however not find any reference that combines the latter or LCL models with SLT. One does also not find any reference that combines LCL models with PIAM/PCS. The present approach fills up these gaps.
The sample of observations is drawn from the 2002 French Parisian travel survey. It considers workers who chose between three available modes of transport for their home-to-work trips: car, public transport, non motorised. The independent variables model travel attributes (costs, time, distances) and individual characteristics (income, location, etc.). Data, results, and derived indicators are discussed in the paper. It is found that the LCL-SLT approach fits data the best.