Abstract
Integration between bike sharing and rail transit provides users with a more flexible travel pattern in an effort to address the “first/last mile” problem. This study aims to examine the determinants influencing the ridership of station-based bike sharing (SBBS) and free-floating bike sharing (FFBS) at rail transit stations. The empirical analysis is based on user transaction records of two bike sharing systems in Nanjing, China. We first apply the k-means cluster method to classify rail transit stations into five types according to the temporal profiles of bike sharing usage for rail transit access. Later, ordinary least squares (OLS) and partial least squares (PLS) regression models are developed respectively by differentiating SBBS and FFBS. Compared with the OLS models, PLS models could address the issue of multi-collinearity and generally have better interpretation abilities. The PLS results reveal that the usage of SBBS for rail transit access shares similarities with FFBS. For example, both of them are positively influenced by population density and the number of restaurants. Meanwhile, different types of rail transit stations exhibit different impacts on the ridership of the two bike sharing systems. Our results show that there exists a substitution effect for rail transit access between two bike sharing systems, that is, SBBS may be more frequently used for commuting trips than FFBS. The findings of this study provide a better understanding of the impact of various factors on the SBBS and FFBS ridership at rail transit stations, thereby helping to promote the integration of rail transit and bike sharing systems.
Acknowledgments
The authors appreciate the Nanjing Bike Sharing Co., Ltd. and Beijing Mobike Technology Co., Ltd. for providing the data used in this study. Thanks to extensive comments from the anonymous reviewers, they have significantly improved the paper.
Notes
1 Note that the transaction records from 0:00 to 6:00 were eliminated since the count of trips in this period is almost negligible. Hence, each rail station has 36 (pickups/returns × 18h) variables for cluster analysis.