Using big data to model travel behavior: applications to vehicle ownership and willingness-to-pay for transit accessibility
MacFarlane, Gregory Stuart
MetadataShow full item record
The transportation community is exploring how new "big'' databases constructed by companies or public administrative agencies can be used to better understand travelers' behaviors and better predict travelers' responses to various transportation policies. This thesis explores how a large targeted marketing database containing information about individuals’ socio-demographic characteristics, current residence attributes, and previous residential locations can be used to investigate research questions related to individuals' transportation preferences and the built environment. The first study examines how household vehicle ownership may be shaped by, or inferred from, previous behavior. Results show that individuals who have previously lived in dense ZIP codes or ZIP codes with more non-automobile commuting options are more likely to own fewer vehicles, all else equal. The second study uses autoregressive models that control for spatial dependence, correlation, and endogeneity to investigate whether investments in public transit infrastructure are associated with higher home values. Results show that willingness-to-pay estimates obtained from the general spatial Durbin model are less certain than comparable estimates obtained through ordinary least squares. The final study develops an empirical framework to examine a housing market's resilience to price volatility as a function of transportation accessibility. Two key modeling frameworks are considered. The first uses a spatial autoregressive model to investigate the relationship between a home's value, appreciation, and price stability while controlling for endogenous missing regressors. The second uses a latent class model that considers all these attributes simultaneously, but cannot control for endogeneity.