Nonlinear effect of accessibility on car ownership in Beijing: Pedestrian-scale neighborhood planning

Abstract

Using a 2017 travel survey dataset and crawled heatmaps and point-of-interests (POIs) data in Beijing, China, this study adopts a gradient boosting decision trees (GBDT) algorithm to measure the relative importance and nonlinear effects of local accessibility, regional accessibility, and transit access on household car ownership. Results show that local accessibility measures such as retail and service density and job density play a more critical role in predicting auto ownership than transit access, while regional access to city centers is the least important. Thus, for reducing car ownership, planning efforts should emphasize improving local accessibility through planning pedestrian-scale neighborhoods (i.e., life-circles). Moreover, nonlinear associations between accessibility measures and car ownership are common. The results suggest that within the 15-min neighborhood life circle, there should be 65–145 retail and service facilities per km2 and block size should be within 150–200 m. Furthermore, residential neighborhoods should be within 400 m of bus stops and 1200 m of metro stations. These findings provide meaningful policy implications for planning pedestrian-scale neighborhoods recently advocated in Chinese cities.