Causal analysis from primary crashes to secondary crashes on freeways using an integrated Bayesian and path...

Secondary crashes on freeways pose substantial safety risks due to the propagation of traffic disturbances following primary crashes. To systematically investigate how primary crashes affect secondary crash occurrence, this study develops a multi-source data fusion framework integrating crash data, traffic flow data, and roadway geometry data. Crash data were obtained from the California Statewide Integrated Traffic Records System (SWITRS), traffic flow variables were extracted from PeMS detector data, and roadway geometric attributes were derived from OpenStreetMap. A preliminary spatiotemporal threshold method was first applied to identify potential primary and secondary crash pairs, followed by a refined identification procedure based on speed deviation patterns across upstream and downstream detectors. Methodologically, this study proposes an integrated Multi-TAN and Object-Oriented Bayesian Network (OOBN) framework to capture both local dependencies among grouped variables and global causal relationships across hierarchical factors. The model incorporates crash conditions, unsafe driving behaviors, crash features, traffic flow characteristics, and environmental factors. Compared with standard TAN and Naïve Bayes models, the proposed Multi-TAN model achieves higher classification accuracy across different secondary crash types, demonstrating improved predictive performance and structural interpretability. Furthermore, mutual information analysis and Top-K path exploration are employed to identify critical risk factors and causal transmission paths. The results indicate that secondary crashes are generated by the nonlinear interaction of multiple factors rather than by a single cause. Multi-vehicle involvement and crash severity are key event-driven triggers, while traffic flow instability, represented by significant changes in volume, speed, and occupancy, forms an important cumulative risk mechanism. Unsafe speed, alcohol involvement, lighting conditions, roadway curvature, and traffic control devices further aggravate the evolution of secondary crash risk. The causal chain analysis reveals a dual-risk structure: high-impact but low-frequency paths dominated by crash severity and multi-vehicle involvement, and lower-intensity but high-frequency paths driven by unstable traffic flow. These findings provide a causality-oriented basis for secondary crash prevention, real-time risk assessment, and targeted freeway traffic safety management.