STRUCTURAL ISSUES AND SOURCES OF BIAS IN RESIDENTIAL LOCATION AND TRAVEL MODE CHOICE MODELS

The chief contribution of this dissertation is to improve the state of the art in applied discrete choice demand modeling in urban economics.; McFadden's Generalized Extreme Value (GEV) model is extended. A set of GEV models which fully relax the Independence from Irrelevant Alternatives (IIA) property of multinomial logit (MNL) models and are approximations of a certain subset of generalized multinomial probit (MNP) models is derived. These GEV models are estimated in the travel mode choice context and the results are compared to the MNL and MNP models. The results show that the estimated parameters are quite comparable. The GEV model is superior to the MNP in that its computational cost is far less and its analytical form and elasticity measures can be expressed in closed form.; Multidimensional extension of the GEV models are also derived in the housing/travel mode choice context. This new model structure is superior to the current nested logit model structure in the sense that it is able to capture the similarity indices among mode alternatives and housing alternatives simultaneously.; Models of joint residential location and travel mode choice are theoretically derived and empirically estimated from aggregate (relative frequency) data based on spatial aggregation units. This dissertation is the first piece of research aiming at the empirical development of large scale aggregate residential demand models. The models are behaviorally valid and yield results which are highly comparable to those of dissaggregate models estimated by others.; Various sources of bias important to discrete choice modeling are identified and empirically analyzed. The sources of bias that are thoroughly discussed and empirically examined include biases due to choice of model type (MNP, GEV, MNL), model structure (simultaneous, sequential), utility functional form, inclusion/exclusion of attributes, measurement errors, sampling errors, and sample size. The relative importance of such sources of bias are empirically identified by the criteria of likelihood ratio statistics, coefficient magnitudes, point estimates of the demand elasticities of housing price, travel cost and travel time and ranges of these demand elasticities. Empirical evidence showed that marginally speaking, biases stemming from model type specification and measurement errors are minor compared to the impact of the other sources of bias. Sampling error, utility functional form bias and model structure bias (simultaneous versus sequential) appear to be more important than missing attribute bias and aggregation bias given that the choice model is reasonably well specified and the data is aggregated on traffic zones.; The dissertation also provides new knowledge about the elasticities of housing rents, travel time, travel costs, and other explanatory attributes of interest to urban economists.