Hybrid dimension reduction and logit models for glare-induced crash severity

Area of Science:

• Traffic Safety
• Transportation Engineering
• Human Factors in Driving

Background:

• Sun glare (sunrise/sunset) and headlight glare at night impair vision, reducing contrast sensitivity and hazard detection.
• These visual impairments create high-risk driving conditions, contributing to severe crashes in specific roadway and lighting contexts.

Purpose of the Study:

• To examine unobserved heterogeneity in glare-induced crash severity using a novel framework.
• To identify distinct crash typologies related to glare and analyze their specific injury severity outcomes.

Main Methods:

• A two-stage, cluster-driven discrete choice framework was developed.
• Cluster Correspondence Analysis (CCA) identified three glare-related crash typologies.
• Cluster-specific multinomial and random-parameter logit models assessed injury severity.

Main Results:

• Crash severity mechanisms are significantly dependent on identified glare-related typologies.
• Low-speed urban angle crashes have lower severe injury risk, especially in daylight.
• High-speed crashes and rural nighttime crashes on unlit roads show higher severe injury likelihood.

Conclusions:

• Integrating clustering with random-parameter logit modeling enhances understanding of glare-induced crash severity.
• Findings support targeted countermeasures: dynamic speed management, improved lighting, adaptive headlamps, and driver education.
• Specific interventions should address the unique risks of different glare-related crash typologies.