Unraveling the mode substitution of dockless bike-sharing systems and its determinants: A trip level data-driven interpretation

Understanding the mode substitution of shared micro-mobility systems is essential for assessing their societal and environmental impact and developing improvement planning instruments. This study carries out a fine-grained analysis of the mode substitution of dockless bike sharing (DLBS) in relation to other transport modes at the trip level, leveraging multi-modal route planning techniques, transaction data of bike-sharing, and travel behavior modeling. More importantly, the study leverages interpretable machine learning to reveal the complex effects of built environment factors on the mode substitution patterns of DLBS based on multiple data sources. The results indicate that the probabilities of DLBS replacing other transport modes present large heterogeneity among different trips and in different urban contexts, which can be successfully quantified by the proposed approach at the trip level. The average substitution rates of bike-sharing to bus, metro, walking and ride-hailing in Shanghai are estimated to be 0.356, 0.116, 0.347 and 0.181, respectively. Built environment factors such as presence of transit systems can explain the variations in the substitution rates of DLBS to a certain transport mode in different urban contexts. Especially, the effects of some built environment factors show complex nonlinear and threshold patterns revealed by the data-driven method. The effects of key built environment factors are quantitatively interpreted and their practical implications discussed.

Automated summary

Understanding the mode substitution of dockless bike sharing (DLBS) in relation to other transport modes is crucial for assessing their impact and planning for improvements. This study utilizes multi-modal route planning techniques, transaction data of bike-sharing, and travel behavior modeling to analyze the mode substitution of DLBS at the trip level. The study also employs interpretable machine learning to uncover the effects of built environment factors on mode substitution patterns. The results show heterogeneity in the probabilities of DLBS replacing different transport modes, and built environment factors play a role in explaining these variations.