Applying a Bayesian multivariate spatio-temporal interaction model based approach to rank sites with promise using...

Area of Science:

• Road Safety Engineering
• Statistical Modeling
• Traffic Management

Background:

• Effective identification of high-risk roadway sites is crucial for targeted safety improvements.
• Existing ranking methods may not fully capture complex spatio-temporal crash dynamics.

Purpose of the Study:

• To develop and apply a Bayesian multivariate spatio-temporal interaction model for ranking roadway segments.
• To compare the proposed ranking approach with traditional and alternative Bayesian methods.

Main Methods:

• Utilized a Bayesian multivariate spatio-temporal interaction model.
• Employed severity-weighted crash frequency and crash rate as decision parameters.
• Applied posterior expected rank and posterior mean as statistical criteria.
• Implemented the model using WinBUGS for analysis of Kaiyang Freeway data.

Main Results:

• The proposed model demonstrated improved goodness-of-fit compared to a multivariate Poisson-lognormal model.
• Incorporating spatio-temporal correlations significantly influenced identified crash factors and safety effects.
• Ranking results differed notably from naïve and multivariate Poisson-lognormal Bayesian approaches.
• Posterior mean and posterior expected rank criteria yielded consistent ranking outcomes.

Conclusions:

• The Bayesian spatio-temporal interaction model offers a more robust approach to ranking roadway sites for safety improvements.
• This advanced modeling technique provides a better understanding of crash-contributing factors and safety effects.
• The findings highlight the importance of considering spatio-temporal dynamics in road safety analysis.