Existing utility-based models of complex choice behavior do not adequately deal with the interdependencies of chained choices. In this paper, we introduce a model of multi-purpose shopping which is aimed at overcoming this shortcoming. In the proposed model, dependencies between choices within as weil as between trips are covered by a recursive definition of trip utility. The standard log-likelihood estimation procedure is used to calibrate the model. Simulation experiments show that estimation results are satisfactorily accurate and robust. Comparison of the model to a conventional choice model using simulated data indicates that eren low tendencies to make multi-purpose trips have a significant influence on predicted destination choice. Furthermore, it is shown that conventional models do not satisfactorily predict simulated multi-purpose behavior.
INTRODUCTIONDemand analysis continues to be a research area in geography and urban planning important for a broad range of issues pertaining to consumer behavior (Golledge and Timmermans 1990; Timmermans and Golledge 1990). The most widely used model in demand analysis is the multinomial logit model, which predicts the probability that consumers will choose a particular alternative, given its locational and non-locational attributes, the attributes of its competitors, and possibly a set of socio-economic characteristics of the consumers. The model is easy to use, is based on sound economic principles, and appears to be rather robust (Borgers and Timmermans 1987).Recent research however has raised a number of reasons for doubting the ability of the multinomial logit model to predict consumer choice in complex situations involving interrelated choices. Although it is possible to use a proxy of the magnitude of multi-purpose shopping as a predictor, the multinomial logit model is typically a model of single-stop, single-purpose choice behavior: the effects of pastor future choices on current choice, and the fact that a trip may involve different purposes are not explicitly taken into account. This suggests that the multinomial logit model may be a weak predictor of multistop, multi-purpose shopping behavior. Because empirical evidence suggests