Automobile Ownership Model That Incorporates Captivity and Proximate Covariance

The modeling effort in this paper is distinguished from previous models of automobile ownership primarily by the use of the dogit ordered generalized extreme value (DOGEV) model rather than the commonly used multinomial logit and ordered probit or logit models. In comparison with other models, the DOGEV model has two distinct features. First, it recognizes the ordinal nature of automobile ownership levels (zero cars, one car, two cars, and three or more cars) by allowing those levels to be correlated in close proximity (i.e., proximate covariance: ownership levels that are close to each other in the ordering have error terms that are correlated). Second, the DOGEV model allows a household’s automobile ownership choice to be captive or constrained to a particular automobile ownership level and, therefore, avoids the potential misspecification of the choice sets for individual households. The modeling approach is based on a behavioral analysis that explains the factors that influence household automobile ownership decisions in the New York City area, a highly urbanized environment. The estimation results uncover the sensitivity of household automobile ownership choices to transit accessibility, urban forms, traffic congestion, parking costs and availability, and the level of access to opportunity sites through nonmotorized transportation. The results also show that New York City data of automobile ownership are well analyzed by the DOGEV model. Particularly, evidence of captivity and ordering (proximate covariance) in the choice set may suggest an additional source of misspecification in the existing automobile ownership literature.