Using multivariate adaptive regression splining (MARS) to identify factors affecting the performance of dock-based bikesharing: The case of Chicago’s Divvy system

This study explores factors contributing to the uneven success of past expansions of dock-based public bikesharing systems, in which middle- and - higher-income neighborhoods have tended to benefit considerably more than poorer neighborhoods. After a review of the differing performance of the three phases of expansion by Chicago's Divvy bikeshare system, this study uses multivariate adaptive regression splining (MARS) to select among more than 100 community- and station-level factors to explain variations in Divvy system usage at the station level. MARS demonstrates that neighborhood racial and ethnic diversity, proportion of condominium units, and job accessibility to public transit are strongly and positively correlated with total annual station trips, whereas percentage unemployed, average distance to Divvy stations, and percentage of residential foreclosures are negatively correlated. Model results are compared with those of earlier studies to foster insights into ways to more accurately predict the use of bikesharing systems across urban neighborhoods.