Modeling injury severity in embankment-related roadway departure crashes

Embankment-related roadway departure crashes pose significant safety risks due to slope geometry, vehicle instability, and elevated impact forces, yet limited research addresses how injury mechanisms vary across roadway environments and over time. This study examines embankment-related crash severity using Texas Crash Records Information System (CRIS) data from 2021 to 2024, employing a hybrid framework combining Random Parameters Logit models with Heterogeneity in Means (RPLHM), partially constrained temporal stability testing, and Natural Language Processing (NLP) based narrative analysis. The framework captures persistent and time-varying effects of roadway geometry, environmental conditions, crash characteristics, and driver attributes across three severity outcomes: no injury, possible/non-incapacitating injury, and fatal/incapacitating injury. Results reveal that roadway alignment, lighting conditions, fixed-object involvement, and occupant protection consistently shape severity, while heterogeneity in geometric and environmental factors indicates strong context-dependent risk patterns. Curved alignments and certain roadway environments increase the likelihood of severe injury, whereas seatbelt use substantially increases the probability of non-injury, though its protective effect varies across contexts. Temporal analyses show that while several determinants remain stable, selected parameters exhibit year-specific variation. Narrative topic modeling highlights recurring mechanisms involving loss of control, slope interaction, and environmental influences. These findings underscore that uniform countermeasures are insufficient and emphasize, the need for context-specific roadside design, slope treatment, speed management, and occupant protection strategies.