A full Bayes multivariate intervention model with random parameters among matched pairs for before-after safety...

Area of Science:

• Traffic Safety Engineering
• Transportation Planning
• Statistical Modeling

Background:

• Intersection safety is a critical concern in urban areas.
• Evaluating the effectiveness of implemented safety countermeasures is essential for informed decision-making.
• Existing methods may not fully account for site-specific variations and correlations.

Purpose of the Study:

• To assess the safety performance of improved intersections in Greater Vancouver.
• To quantify the effectiveness of implemented safety countermeasures using a robust statistical approach.
• To explore the impact of specific design features on crash reduction.

Main Methods:

• Utilized a full Bayes approach with a before-after design and matched comparison groups.
• Employed a multivariate Poisson-lognormal intervention model to analyze crash counts by severity.
• Extended the model with random parameters to address correlations between paired sites.

Main Results:

• Overall, significant reductions in predicted crash counts were observed: 23% for injury and fatality (I+F) and 15% for property damage only (PDO).
• Credible intervals at the 0.95 confidence level were (12%, 33%) for I+F and (6%, 24%) for PDO.
• Specific countermeasures showed varying effectiveness: signal visibility (29% I+F, 21% PDO), left turn phase improvement (15% I+F, 4% PDO), and left turn lane installation (21% I+F, 20% PDO).

Conclusions:

• The full Bayes approach with matched comparison groups improves model fit and reduces variation estimates.
• Extended models effectively account for unobserved heterogeneity in road geometry, traffic, environment, and driver behavior.
• Safety countermeasures demonstrate effectiveness, but impact varies by crash severity and location, necessitating tailored solutions.