Determinants of urban ridesplitting service performance : a spatio-temporal exploration and simulation with big data

Abstract

Ridesplitting is a shared mobility service matching passengers traveling in similar directions, promising reduced fleet size, congestion, and emissions. Despite much attention on improving ridesplitting performance, a comprehensive understanding of performance determinants is lacking, particularly from a spatio-temporal perspective. This thesis fills this gap by exploring ridesplitting performance and its determinants using big data and simulation methods. The research objectives are threefold: 1) to identify the determinants of individuals’ Willingness To Share (WTS) rides; 2) to uncover the factors that impact the Ability To Share (ATS) successfully; 3) to evaluate the spatio-temporal performance improvements of ridesplitting. To meet these objectives, three studies are conducted in Manhattan, New York City, based on extensive trip data from transportation network companies, incorporating both solo and shared trips. The first study uses a big data approach to comprehensively model the determinants of WTS at the individual level. The second study proposes a path-based weighted ATS model to model the ATS at the path level, using the XGBoost-based SHapley Additive exPlanations (SHAP) model for solving and interpreting the model. The third study uses an agent-based simulation model to simulate the performances of ridesplitting service across various scenarios, incorporating the findings from the first two studies. The findings reveal a heterogeneous spatio-temporal pattern of WTS, with the most important determinant being the trade-off between saved trip fare and delayed trip time. The value of time associated with this trade-off is approximately $35/h, $33/h, $31/h, and $28/h during the midnight, morning, afternoon, and evening periods respectively. A larger ratio between the saved trip fare and delayed trip time is more favourable for ridesplitting. The impact of sociodemographic and built environment variables on WTS is also analysed. The determinants of ATS are identified using the proposed path-based weighted ATS model. The most crucial determinant for ATS is trip intensity, particularly for medium length trips, followed by short-distance, and then long-distance trips. Road network characteristics also influence ATS, with higher road speeds contributing to better ATS, while over-fragmented blocks and one-way dominated road networks adversely impact ATS. A comparison between ridesplitting and solo trips shows that ridesplitting can serve more passengers, reduce wait time, and increase vehicle occupancy. If all passengers adopt ridesplitting, fleet size could potentially be reduced by 32.8%. Meanwhile, ridesplitting mainly enhances the service experience in bustling urban centre areas and at peak times, with less significant improvements in more distant areas and non-peak times. Based on these findings, policy implications are offered. For successful development, strategies include increasing WTS by lowering fares and limiting travel delays, and creating attractive and efficient environments for sharing. To avoid unexpected problems, strategies for healthy development include improving the spatial accessibility of ridesplitting service and paying close attention to the primary user group. These approaches and results can guide future studies on ridesplitting performance determinants, and inform policies to develop this promising mobility service for sustainable transportation.