Exploring the causal effects of hazardous driving behaviors on pedestrian-vehicle crash injury severity based on causal machine learning.

Pedestrians usually sustain more severe injuries in traffic crashes due to the lack of protection. Although previous studies have analyzed crash-influencing factors, the causal relationships between injury severity and factors remain limited. Thus, the study aims to investigate the causal effects of factors, particularly the hazardous driving behaviors, on pedestrian injury severity. It employs causal machine learning and the Shapley Additive exPlanations to analyze the marginal contribution of the feature variables on crash injury severity. Furthermore, the causal tree model is used to reveal the heterogeneous causal effects. The results indicate that 1) the pedestrian injury severity is influenced by various factors, such as pedestrian age, speed limit, and vehicle type, 2) hazardous driving behaviors can deteriorate pedestrian injury severity, with 'failing to yield right of way' showing the highest mean conditional average treatment effect (CATE) (0.768), followed by 'careless driving' (0.626) and 'violating traffic signs or signals' (0.590), and 3) hazardous driving behaviors, road speed limits, and pedestrian age jointly contribute to the pedestrian injury severity. The study reveals heterogeneity in the causal relationship between hazardous driving behaviors and the injury severity of pedestrian-vehicle crashes, which serves to develop differential intervention strategies to mitigate pedestrian injury.