Download files
Abstract
Against the backdrop of dynamic socio-technical change, travel demand analysis strives to derive insights into travel behaviour from new and established data sources to support strategic and operational processes in the public and private sectors. A principal concern of current travel demand analysis is the representation of unobserved heterogeneity to understand the complex patterns of travel demand. Hierarchical Bayesian models promise to satisfy the desiderata of contemporary travel demand analysis. However, they are not widely used in travel demand analysis, and existing estimation methods do not suit contemporary inference problems. This thesis has two aims: i) to leverage the hierarchical Bayesian modelling paradigm to accommodate flexible representations of unobserved heterogeneity in travel demand models and to demonstrate the practical value of these representations in empirical applications, and ii) to advance and benchmark computational methods for posterior inference in hierarchical Bayesian models of travel demand. The aims are addressed over four main chapters. First, a hierarchical Bayesian multivariate Poisson log-normal model is used to analyse intergenerational differences in transport mode use among young adults in Germany. Second, parametric and semiparametric representations of unobserved heterogeneity in mixed logit models are compared and leveraged to infer distributions of willingness to pay for features of shared automated vehicle services in New York City. Third, several variational Bayes (VB) methods for posterior inference in mixed logit models are extended to admit utility specifications with both fixed and distributed parameters. Fourth, a VB method for posterior inference in mixed logit models with unobserved inter- and intra-individual heterogeneity is derived. In both the third and fourth main chapters, simulation studies are used to benchmark the VB methods against Markov chain Monte Carlo methods and frequentist maximum simulated likelihood estimation. As a whole, this thesis exhibits that the hierarchical Bayesian modelling paradigm constitutes a powerful and flexible approach for the formulation of disaggregate, behavioural models of travel demand. Moreover, the empirical applications demonstrate that substantive, policy-relevant behavioural insights can be derived using hierarchical Bayesian models. The thesis also lays the foundations for hierarchical Bayesian models of travel demand to be scaled to massive datasets.