Exploring Model Uncertainty and Behavior Uncertainty - A Temporal Transferability Assessment of Vehicle Ownership Models for Boston Metropolitan Area

In this study, the authors analyze model uncertainty and behavior uncertainty in vehicle ownership choice models. The authors use temporal transferability tests to compare alternative model structures − logit (MNL), ordered logit (OL) and nested logit (NL), and different model specifications. The authors also test preference stability. Using data from Boston Metropolitan Area in 1991 and 2010, it is found that OL is more temporally transferable than MNL and NL, although its local error is much higher than MNL or NL. As model specification becomes more complex, both local error and transfer error decrease, but only to a certain degree; and the decrease in transfer error is smaller than local error. The authors find preferences are in general not stable over time. Preference parameters for local built environment factors and transit access are stable, while the effects of regional location factor and most socio-economic and demographic factors have evolved. If it is assumed preference had not changed, one would significantly under-predict 0-vehicle households by 42.5%, and over-predict 2-vehicle households by 14.8% in population forecast for 2010.