Exploring Model and Behavior Uncertainty: Temporal Transferability Assessment of Vehicle Ownership Models for Boston, Massachusetts, Metropolitan Area

This paper examines model and behavior uncertainty in vehicle ownership choice models. Temporal transferability tests are used to compare alternative model structures—multinomial logit (MNL), ordered logit (OL), and nested logit (NL)—and different model specifications. Preference stability is also tested. Through the use of data from the Boston, Massachusetts, metropolitan area in 1991 and 2010, it is shown that OL is more temporally transferable than MNL and NL but has a much higher local error. As the model specification becomes more complex, local error and transfer error decrease to a certain degree; the decrease in transfer error is smaller than that of local error. Overall, household vehicle ownership preferences are not stable over time but vary by specific parameter estimates. The relationships between vehicle ownership and local built environment factors and transit access are stable; the relationships related to regional location and most socioeconomic and demographic characteristics have evolved. Under an assumption that preferences have not changed over time, the use of 1991 models to forecast household vehicle ownership in 2010 underpredicts zero-vehicle households by 42.5% and overpredicts two-vehicle households by 14.8%.