Exploring the relationship between ride-sourcing services and vehicle ownership, using both inferential and machine learning approaches

Ride-sourcing services are getting more popular each year, and their markets are growing. Much has been speculated, but not much has been tested regarding the impacts of ride-sourcing services on the transportation system. In this study, we examine the relationship between ride-sourcing services and vehicle ownership of households, by using the most up-to-date (2017) national household travel survey data. To better capture the effect of ride-sourcing services on vehicle ownership, we controlled for the effect of socioeconomic characteristics of households and built environment variables, i.e., density, diversity, design, and distance to transit. Two approaches were used to model vehicle ownership: a probabilistic or inferential model (i.e., multilevel Poisson), and a machine learning method (i.e., random forest). This is the first study to utilize such advanced methods to model vehicle ownership and capture non-linear relationships, using the largest sample of household travel records ever assembled for such a study. The results suggest that there is a negative correlation between using ride-sourcing services and vehicle ownership. Vehicle ownership is also negatively associated with the number of years Uber, as the biggest ride-sourcing service, has operated in a county. The relative contributions of ride-sourcing variables, however, are very limited compared to other variables controlled in this study which makes intuitive sense. For urban planning and design practices, this study suggests that the probability of car shedding will increase if the usage of ride-sourcing services becomes a habit, these services become more available, and built environments become more dense, connected, and transit-served.