A Heterogeneous Effects Analysis Method of Highway Crash Factors Based on Causal Framework

The analysis of heterogeneous effects on traffic crashes is crucial for understanding their causal mechanisms and enhancing targeted safety management strategies. However, current methodologies for modeling crash heterogeneous effects lack smooth methods for selecting optimal controls. This study proposes an intuitive variable selection method to improve heterogeneity analysis of crash data, as well as performance evaluation and validation tests. The method utilizes causal discovery algorithms to obtain causal diagrams for selecting confounders, moderators, and neutral control factors in observational collision data. The effectiveness and performance of these methods are assessed through the quality of Heterogeneous Treatment Effects (HTE) estimation. Using a real-world highway crash data, the proposed variable selection process based on causal framework is illustrated. Results indicate that most HTE estimation models perform well in terms of goodness-of-fit and robustness when employing the graphical variable selection method. Notably, models based on adjusted causal diagram and forest-based double-robust learning estimators perform the best across all model. This approach overcomes the challenges in selecting control and moderator factors in crash heterogeneity analysis, leading to more accurate effect estimation results. This study contributes to encouraging discussions on the causal pathways of crash occurrence and provides recommendations for optimizing road safety modeling, analysis and management in the field of crash analysis.