Carpool to work: Determinants at the county-level in the United States

Abstract

Using US county-level data for 2017, this paper adopts a data-driven approach to study the main factors influencing carpooling for home-to-work trips. The potential explanatory variables quantify demographic, situational and judgmental characteristics of the counties. Since the proportion of workers carpooling for their commute is not (spatially) randomly distributed at the county level, the inclusion of spatial effects improves considerably the model's fit. The Spatial Autoregressive models show that eight variables (four related to demographics, three to situational and one to judgmental) do the best job of explaining the rates of carpooling. A Spatial Quantile Autoregression is further applied as a flexible and interpretable method to address the fact that some of the leading variables have varying effects on counties with different levels of carpooling. For instance, our results suggest that the agglomeration effect, measured by an increase in population density, has a gradual change in trend pattern because it encourages ridesharing among counties with low levels of carpooling, whereas it deters shared trips in high intensity carpooling areas. Alternatively, a decrease in car ownership, a variable strongly associated to counties' income, will lead to the largest increase in employees using carpooling for their home-to-work travels, and this relationship do not vary across quantiles.