Ineffective built environment interventions: how to reduce driving in American suburbs

Keywords: land use, travel behavior, nonlinear relationship, machine learning, heterogeneity
Abstract Type: Paper Abstract

Abstract

Why do we single out American suburbs while examining the relationship between the built environment and auto use? First, most population growth will occur in the suburbs because of limited developable lands and high development costs in urban areas. Second, suburbanites on average drive much more miles than their urban counterparts. Third, most studies use data from a metropolitan area as a whole and overlook the differing responses of urban and suburban residents to the built environment. Therefore, designing effective built environment interventions in the suburbs is key to sustainable transportation.

Using household travel survey data from the Twin Cities area, USA, we apply the gradient boosting decision tree method to address two research questions: 1) Are built environment attributes equally important to predicting driving distance of residents living in urban and suburban areas? 2) How do the relationships between built environment attributes and driving distance in suburban areas differ from those in urban areas?

The results show that suburban residents are less sensitive to the built environment than urban residents. More importantly, built environment policies that work in urban areas might not be useful in suburban areas. Many studies have proposed population densification and mixed-use development to mitigate auto use. However, this study suggests that although promising in urban areas, these policies are ineffective in suburban areas. Instead, promoting job accessibility and densifying intersection density appears promising to reduce auto use in suburban areas. Increasing transit supply has a small but nontrivial contribution to mitigating auto use.