In recent years there has been a considerable growth of interest in exploring the application of reference-dependent choice models in transport demand analysis. These choice models generalise the standard utility maximising model by assuming that individuals evaluate potential alternatives in terms of the gains or losses they offer relative to some assumed reference point. If such reference point based decision processes exist within the population, then they have important implications for a number of aspects of demand modelling, most particularly for the computation of willingness to pay measures and for associated project appraisal applications. They also inject an extra and potentially highly significant source of hitherto unobserved heterogeneity into decision making. For these and related reasonsit is therefore important that the nature and extent of such reference point effects be identified more fully than at present. Reference-dependent choice models have shown some promise in particular applications but existing applications have also raised a number of important issues, which to date remain unresolved. The aim of this paper is to address two of these issues, both of which relate to the definition of the reference point. The first issue is whether this reference point should be defined in attribute space (i.e., as a particular "reference alternative") or in utility space (i.e., as a "reference level of utility", which will in general correspondto many possible alternatives). The theoretical and practical implications of these two interpretations are quite different. The second issue concerns the definition of the reference point itself. In most existing applications it is defined as some notion of 'current behaviour', so that gains and losses are in effect measured relative to this current behaviour. However, this seems overly restrictive, especially when considered in the context of the various ambiguities associated with the concept of 'current behaviour' itself. Moreover, a plausible behavioural case can be made for a number of alternative concepts of reference point, such as 'desired behaviour' or 'expected behaviour' and for the importance of inter-personal or social comparisons. These issues are impossible to resolve by theoretical argument alone and this study explored them empirically using a number of mode choice stated preference datasets. Models in which the reference points are treated as missing data and estimated endogenously (i.e., simultaneously with other taste parameters) using a version of the expectation maximisation (EM) algorithm were used. The results from these models are comparedwith the results from conventionally specified utility maximising models.For the covering abstract see ITRD E145999
Samenvatting