Modelling injury severity in single-vehicle crashes using full Bayesian random parameters multinomial approach

Area of Science:

• Transportation Engineering
• Traffic Safety Analysis
• Statistical Modeling

Background:

• Single-vehicle (SV) crash severity modeling traditionally considers spatiotemporal correlations but often overlooks spatiotemporal interactions.
• Existing models may not fully capture the complex unobserved heterogeneity and dynamic relationships influencing crash outcomes.

Purpose of the Study:

• To propose a novel statistical approach, the spatiotemporal interaction correlated random parameters logit with heterogeneity in means and variances (STICRP-logit-HMV), for enhanced SV crash severity modeling.
• To systematically characterize unobserved heterogeneity, spatiotemporal correlations, and spatiotemporal interactions in SV crashes.
• To compare the performance of the proposed STICRP-logit-HMV model against several established statistical frameworks.

Main Methods:

• Developed four flexible interaction formulations (linear, Kronecker product, mixture-2, mixture-5 models) to uncover spatiotemporal interactions.
• Established and compared four candidate models: random parameters logit (RP-logit), RP-logit with heterogeneity in means and variances (RP-logit-HMV), correlated RP-logit-HMV (CRP-logit-HMV), and spatiotemporal CRP-logit-HMV (STCRP-logit-HMV).
• Utilized single-vehicle crash data from Shandong Province, China, for model calibration and validation.

Main Results:

• The STICRP-logit-HMV model demonstrated superior performance compared to all benchmark models, confirming the importance of incorporating spatiotemporal interactions.
• The RP-logit-HMV model outperformed the basic RP-logit, highlighting the benefit of accounting for heterogeneity in means and variances.
• The STCRP-logit-HMV model showed improvement over CRP-logit-HMV, indicating the value of addressing spatiotemporal crash mechanisms.
• Among interaction formulations, the mixture-5 component within the STICRP-logit-HMV model yielded the best model fit.

Conclusions:

• The STICRP-logit-HMV model provides a robust statistical framework for analyzing SV crash severity, outperforming existing methods by comprehensively modeling spatiotemporal correlations and interactions.
• Factors such as young drivers, male drivers, and non-dry road surfaces exhibit significant heterogeneity effects on crash severity.
• The findings offer valuable insights for transportation professionals to develop more effective traffic safety strategies and improve road safety outcomes.