Abstract
Accurately modeling crash severity on rural two-lane roads is essential for effective safety management, yet standard single-level approaches often overlook unobserved heterogeneity across road segments. This study analyzes 19,956 crash records from 99 rural roads in Iran, incorporating crash-level predictors and road-level covariates. Three binary logistic frameworks are compared: a single-level generalized linear model, a multilevel model with a random intercept, and a multilevel model with random coefficients. The random-coefficient model achieves the best fit and substantially improves predictive performance: classification accuracy rises from 0.62 to 0.71, recall from 0.32 to 0.63, and AUC from 0.570 to 0.775. Simulation results reveal notable variability in pavement and lighting effects, demonstrating how local context influences crash risk and targeted safety interventions.
Interactive research sketch
Explore the mechanism
Conceptual interaction based on the study theme—not a reproduction of the reported statistical model.
Select a road segment to reveal contextual variation.
Illustrative only. Consult the paper for methods, assumptions, uncertainty, and validated results.
Access and rights
Copyright-aware discovery
This record links to an openly available paper or preprint through its DOI. Reuse remains governed by the licence stated at the destination.