Endogeneity and Sampling of Alternatives in Spatial Choice Models

Abstract

Addressing the problem of omitted attributes and employing a sampling of alternatives strategy, are two key requirements of practical spatial choice models. The omission of attributes causes endogeneity when the unobserved variables are correlated with the measured variables, precluding the consistent estimation of the model parameters. The consistent estimation while sampling alternatives in non-Logit models has been an open problem for three decades. This dissertation is concerned with both the endogeneity and the sampling of alternatives in non-Logit models, two problems that have hindered the development of suitable modeling tools for urban policy analysis, but have been neglected in spatial choice modeling. For the problem of endogeneity, this research applies, enhances, adapts, and develops efficient and tractable methods to correct and test for it in models of residential location choice, and also develops novel methods to validate the success of the correction. For the problem of sampling of alternatives in non-Logit models, this study develops and demonstrates a novel method to achieve consistency, relative efficiency, and asymptotic normality when the underlying model belongs to the Multivariate Extreme Value class. This development allows for the estimation of spatial choice models with more realistic error structures. Monte Carlo experiments and real data from Lisbon, Portugal, are employed to illustrate the significant benefits of these novel methods in correcting for endogeneity and addressing sampling of alternatives in non-Logit models, with specific reference to urban policy analysis.