Uncovering the Complex Causal Mechanisms of Road Traffic Collisions at Intersections Using Piecewise Structural Equation Modelling

Understanding the causes of traffic collisions is crucial for road designers, engineers, and policymakers to improve road safety at intersections. Design standards aim to minimize the severity and frequency of collisions. However, the factors that may affect traffic collisions are extensive. Their causal mechanisms can be complex, with feedback loops between traffic flows, visibilities, speeds, risk perception, speed limits, and other geometric characteristics of intersections. Structural Equation Modelling (SEM) is commonly used in behavioural sciences to understand complex causal paths, including travel behaviour studies. However, SEMs cannot robustly represent non-normally distributed datasets and rare count events, and little literature exists on their application to road traffic collisions. To address this limitation, this paper proposes a piecewise Structural Equation Modelling (pSEM) technique, which can handle count responses (i.e. number of collisions) to represent the complex causal relationships that lead to collisions. Application of pSEM technique is compared with conventional SEM. The Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) values demonstrate that pSEM is a more robust approach to model collisions at unsignalized intersections than conventional SEM. In terms of prediction ability, AutoML is much more robust than pSEM and SEM. However, due to difficulties of interpretation for AutoML, it is not recommended for policy implications.