The effect of carsharing on vehicle holdings and travel behavior: A propensity score and causal mediation analysis of the San Francisco Bay Area

We examine the impacts of carsharing on travel behavior utilizing a San Francisco Area subsample of the 2010–2012 California Household Travel Survey. We control for self-selection bias due to differences in observed characteristics of the respondents using propensity-score based matching. We find that vehicle holdings of carsharing members are substantially and significantly lower than for non-members with similar characteristics in terms of individual and household demographics and built environment features of both residential and job location. These differences increase as the propensity to enroll in carsharing programs increases. A latent construct, which measures the propensity to own or utility from owning vehicles and rises with numbers of vehicles owned, is lower for members by 0.3–1.3 standard deviations relative to non-members. Members are also likely to walk, bike, and use transit more frequently than non-members. However, these differences are relatively minor and tend to be statistically non-significant. Future research should control for self-selection bias arising from differences in unobserved characteristics of respondents, as well as simultaneity bias whereby decisions concerning vehicle ownership both influence and are influenced by the decision to join carsharing programs.