Multi-level analyses on the nearest-first matching policy of on-demand chauffeured ride-hailing service

On-demand ride-hailing services have received widespread interest from industry and academic. However, very few researches have empirically investigated the matching policy of ride-hailing platforms with real data. This study contributes in developing multi-level regression models to evaluate the nearest-first matching policy with real-world city-wide data collected from a ride-hailing platform UCAR in China. Specifically, after observing spatial-temporal patterns and the inherently hierarchical structure of the data, we propose multi-level logistics regression models for separately predicting the matching probabilities of passengers' trip requests in the off-peak, eve-peak, and morning-peak hours, where trip requests are grouped by the origins and destinations with cross-classification. The intercepts and slopes in the models can vary with the groups. The multi-level models are verified to be more appropriate and accurate than the traditional logistics regression models. The multi-level models can identify heterogeneous effects of the predicting variables on the matching policy among the groups and allow the discovery of many new findings on matching policy. These findings provide valuable policy implications for ride-hailing platforms to design an improving matching policy.