Individual Versus Grouped SP Models -some Experiments
KROES E, Hague Consulting Group, The Netherlands and ClRILLO C, FUNDP, Belgium
The subject of taste variation, the fact that different individuals may have different preferences and hence different utility co-efficients, has not received much attention in the SP literature. In traditional marketing research, models were estimated fo
The subject of taste variation, the fact that different individuals may have different preferences and hence different utility co-efficients, has not received much attention in the SP literature. In traditional marketing research, models were estimated for each respondent separately, resulting in separate co- efficients for each individual (eg. Kroes and Sheldon 1988). This dealt effectively with the taste variation issue, but required the estimation of extremely large numbers of co-efficients, and resulted in problems of low degrees of freedom (Bates 1988). As a logical consequence, the individual- level co-efficients are bound to be rather inaccurate.
In transport applications, SP models tend to be estimated at group-level, for segments of individuals who are assumed to have similar tastes. This greatly reduces the number of co-efficients to be estimated, but requires a very strong hypothesis concerning the homogeneity of preferences: each individual within the segmentis assumed to have identical preferences and hence utility co- efficients. There is evidence that this hypothesis is unrealistic, and even within seemingly homogenous segments of travellers substantial taste variation can and has be found (eg. Bradley, Ben-Akiva and Bolduc 1992).
The question arises: what is the right level of aggregation for SP model estimation? The aim of our paper is to address this question, and to throw some new light on the issue. We will do this by presenting and discussing evidence that has been available for some time, but which is not widely known. Then we will present and discuss some new material produced by ourselves. We will conclude by providing some recommendations for SP modellers, but also by specifying areas which would benefit from further research.
Association for European Transport