Application of latent variables in transport planning models.

摘要

Latent variables are unobservable and cannot be measured directly. Examples include personality, happiness, and accurate measurement of income. Latent variables have been applied extensively in many fields, such as psychology, health science, education, marketing, and economics. Ignoring latent variables will cause bias estimates and thus damage to model forecasting power. This dissertation makes an effort to extend latent variable application in transport planning models. The application of latent variables is presented in three different ways to capture heterogeneity of individuals, heterogeneity of urban form, and measurement errors by employing appropriate approaches. The first application employs a latent class choice model to estimate the impact of latent lifestyle segments on household location decisions. The second application is to extract urban form factors using a factor analytic method and to examine the role of heterogeneous urban form on car ownership decision across cities. The third application is to correct for hidden measurement error by incorporating indicators in the extended choice model with structural equations.;The results from three empirical studies show that: (1) lifestyle segments exist and they are key determinants of residential location behavior; (2) urban form measurements, such as population density, road infrastructure supply, and urban scale, have significant influence on private car ownership across mega-cities in China; and (3) measurement error can cause serious bias. The major contributions of this dissertation are: (1) it is the first application of latent class choice model estimating lifestyle groups and corresponding preference on residential location simultaneously; (2) it is the first study using both aggregate analysis and disaggregate analysis to investigate urban form impact on private car ownership across mega-cities; (3) it is the first application using the extended choice model framework to correct for measurement error in a mode choice model. The dissertation demonstrates how incorporating latent variables in the transport planning models enhances the behavioral representation of the underlying decision process by linking to the explanatory variables.