Capturing unobserved heterogeneity and temporal variation in discrete choice models have been of interest for researchers in the last two decades. However, understanding how those two aspects might influence individuals’ decisions is still under debate.
This thesis looks at the existing gaps in the literature to be capable of making additional contributions to research in relation to the two areas mentioned above. It is specifically focused on car ownership models for the empirical analysis due to the accessibility of longitudinal data sources. It was thoroughly investigated the appropriateness of the data available to be capable of estimating dynamic effects and controlling for various sources of unobserved heterogeneity. Once the data sources were identified and made accessible to the author, a robust data assembly was undertaken to produce the datasets for modelling. The models estimated in this research were gradually made more complex by adding terms to control for state dependence, random effects, and time-varying coefficients. As a result, six car ownership choice models were further analysed in terms of model performance, forecast accuracy, temporal transferability,
and policy implications.
The analysis of the model results concluded that there is a strong case for considering the addition state dependence of car ownership, and if so, to obtain more stability in the model estimates. In addition, unobserved heterogeneity effects using random effects were found to be significant. Those effects were notably reduced when time-varying coefficients were added for the variable household income and the number of employed adults in the household. The addition of non-standard variables to car ownership models, such as work commute by public transport and childbirth life event, contributed to explain car ownership changes over time.