Exploring the heterogeneous relationship between abnormal driving events and freeways crash risk: A two stage analysis framework integrating causal inference and random parameters logit

Zhou, Mo; Yan, Ying; Wang, Tao; Kieu, Minh; Yuan, Huazhi; Zhang, Changan · 2026 · DuckDuckGo/ScienceDirect

DOI: 10.1016/j.aap.2026.108557

URL: https://doi.org/10.1016/j.aap.2026.108557

archive: indexed pipeline: cataloged

Abstract

Two-stage analytical framework examining abnormal driving events (ADEs) and crash risk on freeways. Stage 1: causal forest with debiased ML (CF-DML) to quantify ADE exposure effects on crash risk. Stage 2: random parameters logit model (RPLHM) for pre-crash ADE distribution patterns. Results: hard acceleration/braking positively associated with crash risk with context-specific heterogeneity; sharp turning correlates with reduced risk. Temperature, traffic volume, truck proportion, and speed dispersion moderate ADE-crash relationship. Three prototypical risk contexts identified: ADE-informative, pre-crash ADE-dense, ADE-sparse.

Access

Route: Elsevier ScienceDirect (paywalled); DOI accessible via institutional subscription

Publisher: Elsevier

Topics