A Severity Analysis of Accidents of Delivery Riders Based on Partial Proportional Odds Model

The existing studies of causal analysis on accidents severity of delivery riders mainly focus on partial fac-tors such as rider, behavior, time, space, road, environment, and accident characteristics. The differences are not quantified in the impacts of different factors on the severity of accidents from the seven aspects above, without con-sidering factors such as takeaway type, number of entrances, angle of intersection, comfort index, and other factors. In addition, when there are both unordered and ordered binary or multi-categorical variables in the independent vari-ables, the established models are limited by the parallel-lines (PL) assumption, and fail to own the flexibility of al-lowing some variables to comply, while others violate this assumption. A total of 1 473 accidents related with deliv-ery riders in Xi'an are selected to analyze the severity distribution and the spatiotemporal distribution. A total of 25 potential influencing factors are selected from the seven aspects. A partial proportional odds (PPO) model is developed to clarify the significant influence of various factors on the rider injury severities in delivery crashes and the vi-olation status of the PL assumption. The corresponding marginal effects are carried out to quantify the differences between and within the contributing factors. The results show that there is a"double peak"phenomenon in the tem-poral distribution of the delivery-involved crashes, and the crash density in urban areas is higher than in suburban ar-eas. The proportion of minor injuries to riders on road sections (35.57%) is higher than at intersections (31.76%). The PPO model performs better than the ordered Logit model and the generalized ordered Logit model. The deliv-ery type, season, location, number of entrances, intersection angle, road surface, weather, and comfort index all fol-low the PL assumption. There are significant differences in the crash severity among different significant factors. Traffic violations such as running red lights, going in the wrong direction, and speeding have the greatest impacts on the rider crash severity, with the maximum absolute value of the marginal effects exceeding 51%. However, some unexplored factors such as number of entrances, intersection angles, bicycle lane, delivery type, and comfort index significantly affect the severity (8% to 37%).