Addressing spatial complexities in residential location choice models

Over the last two decades, there have been limited advances in the conceptualization of, and the modeling methodology for, the residential location choice problem. A widely used methodology for modeling individual household’s residential choice is discrete choice analysis. Analysts typically consider administratively defined zones as discrete choice alternatives and apply the logit models to the residential choice problem in the same manner as for non-spatial contexts. This research argues that there are distinctive features of the residential choice problem that distinguish it from non-spatial choice problems. Failure to account for these features may lead to erroneous analytical results and ineffective spatial policies. Two important spatial features of the residential choice problem are addressed in this study. The first feature relates to the perceived similarity between neighboring choice alternatives that are intangible or difficult to quantify. To address the problem, this dissertation develops the mixed spatially correlated logit (MSCL) model by superimposing a mixing structure to accommodate unobserved heterogeneity across households over a closed form analytic structure that accommodates unobserved inter-alternative correlation. The empirical application of the model shows that the MSCL structure is both conceptually and statistically superior to the conventional modeling approach. The second spatial issue addressed in this dissertation is the representation and measurement of spatial factors. By measuring spatial factors over administratively defined zones, the conventional grouped alternatives approach fails to relate the configuration of spatial units to decision makers’ perception of space. The dissertation proposes a multi-scale structure to replace the conventional ‘flat’ approach. The proposed structure is innovative in that it allows the choice factors’ spatial extent of influence be determined endogenously. In addition, the multi-scale model can be used to test alternative hypothetical representations of neighborhoods as perceived by different households for different residential alternatives. The empirical application of the model demonstrates that social-economic and demographic factors generally have a smaller spatial extent of influence on residential choice than land-use factors. The results also show differing effects of choice factors when different spatial definitions are employed, suggesting the need for future research on behaviorally-realistic spatial representations.