Analyzing the heterogeneous effects of curve alignment and grade on truck crash injury severity on mountainous freeways: A correlated random-parameter binary logit analysis.

OBJECTIVES: To examine how roadway geometry and related crash, vehicle, and environmental factors are associated with truck crash injury severity on mountainous freeways while accounting for unobserved heterogeneity and covariance among random coefficients. METHODS: This study analyzes truck crash data from Yunnan Province, China, from 2015 to 2019. Three binary logit specifications were estimated: a fixed-parameter model, a random-parameter model, and a correlated random-parameter model. Marginal effects were calculated to quantify model-specific average discrete changes in killed and serious injuries (KSI) probability. RESULTS: The correlated random-parameter specification produced the lowest Akaike Information Criterion (AIC), indicating that accounting for unobserved heterogeneity and parameter covariance improves empirical fit. Rear-end collisions, large trucks, curves, non-level grades, and wet surfaces are associated with higher severity, while the guardrail estimate is sensitive to model specification. In contrast, sideswipe crashes, concrete pavement, dry surfaces, winter, summer, afternoon, and evening are associated with lower KSI probabilities. Among the risk-increasing variables in the correlated random-parameter model, non-level grade produced the largest positive average discrete change in KSI probability, while curve alignment also showed a significant positive association. Significant correlations were identified between rear end crashes and non-level grades, and between dry surfaces and summer conditions. CONCLUSIONS: Roadway grade, curve alignment, rear-end crashes, surface condition, and truck type were important conditional correlates of injury severity. The findings provide valuable insights for truck crash prevention and mountainous freeway safety management.