A sample selection model with multinomial endogenous switching: Addressing sample selection bias on binary and continuous outcomes of household vehicle miles traveled

The sample selection modeling approach has been used to quantify the influence of residential self-selection (RSS) on travel behavior outcomes within the framework of sample selection bias (SSB) correction. A limitation of this approach is the reliance on a simple binary endogenous switching mechanism, which oversimplifies residential choices as a binary decision. To address this, this study proposes a sample selection modeling framework that incorporates multinomial endogenous switching. It also accommodates correlated alternatives in multinomial residential choice and handles binary or continuous travel behavior outcomes. We apply this model to data from the 2017 US National Household Travel Survey, which includes 129,587 households. It identifies the SSB influences and the resulting average effects of living in each of the four neighborhood types on household vehicle miles traveled (VMT) for binary and continuous outcomes: (i) whether a household took a car trip on the survey day and (ii) VMT for those households that took a car trip. For the binary outcome, the average predicted probability of taking a car trip is highest in the Second City neighborhoods, followed by the Suburban and the Urban neighborhoods, and lowest in the Small Town/Rural neighborhoods. For the continuous outcome, expected household VMT is highest in the Second City neighborhoods, followed by the Suburban and the Small Town/Rural neighborhoods, and lowest in the Urban neighborhoods. While the identified SSB influences may also capture certain unobserved built environment characteristics, we nevertheless derive valuable insights for the formulation of more effective land-use and transport policies.