Modeling correlation and heterogeneity in crash rates by collision types using full bayesian random parameters...

Area of Science:

• Transportation Engineering
• Traffic Safety Analysis
• Econometrics

Background:

• Road crashes exhibit diverse types, often influenced by unobserved risk factors.
• These factors can induce correlations and heterogeneity in crash rates across different collision types.
• Existing models may not fully capture these complex interdependencies.

Purpose of the Study:

• To propose a novel random parameters multivariate Tobit (RPMV-Tobit) model.
• To evaluate risk factors influencing crash rates for multiple collision types simultaneously.
• To compare the performance of the proposed RPMV-Tobit model against established statistical models.

Main Methods:

• Utilized crash data from 367 freeway diverge areas over three years.
• Focused on three primary collision types: rear-end, sideswipe, and angle collisions.
• Employed a Bayesian framework to compare RPMV-Tobit with multivariate Tobit (MV-Tobit), random effect multivariate Tobit (REMV-Tobit), and independent univariate Tobit (IU-Tobit) models.

Main Results:

• The MV-Tobit model demonstrated superior fit over IU-Tobit models, confirming the benefit of accounting for crash type correlations.
• Both RPMV-Tobit and REMV-Tobit models outperformed the MV-Tobit model, highlighting the importance of addressing unobserved heterogeneity.
• The RPMV-Tobit model showed a greater performance improvement than the REMV-Tobit model.
• Significant heterogeneous effects on crash rates were identified for factors like lane-balanced design, number of lanes, speed limit, and speed difference.

Conclusions:

• The RPMV-Tobit model is a more effective approach for comprehensive crash rate modeling.
• The model enhances traffic safety evaluation by accurately capturing correlations and heterogeneity.
• Findings underscore the need for advanced statistical methods in traffic safety research.