A multivariate spatial crash frequency model for identifying sites with promise based on crash types

Area of Science:

• Transportation Engineering
• Traffic Safety Research
• Statistical Modeling

Background:

• The systemic approach to road safety management aims to identify sites with promise (SWiPs) for targeted interventions.
• While proponents claim superior effectiveness over hot spot approaches, systemic methods face challenges due to low precision in crash frequency models.
• This imprecision arises from analyzing crash subsets (by type), leading to increased variability.

Purpose of the Study:

• To develop a more precise statistical model for road safety analysis.
• To propose a multivariate spatial model for simultaneously modeling crash frequencies across different crash types.
• To enhance the precision of identifying sites with promise (SWiPs) for effective safety interventions.

Main Methods:

• Utilized crash, traffic, and roadway inventory data from rural two-lane highways in Pennsylvania.
• Developed and tested a multivariate spatial model incorporating both multivariate and spatial correlations.
• Compared four models: with and without multivariate and spatial correlations, to assess model fit.

Main Results:

• The multivariate spatial model considering both correlations demonstrated the best fit among the tested models.
• Multivariate correlation between crash types at a site was found to be more influential than spatial correlation between adjacent sites.
• The model successfully enhanced precision in predicting crash frequencies for various crash types.

Conclusions:

• The proposed multivariate spatial model significantly improves the precision of crash frequency prediction in road safety management.
• The integration of multivariate and spatial correlations offers a robust framework for identifying sites with promise (SWiPs).
• Findings highlight the critical role of multivariate correlations in understanding and mitigating diverse crash types.