A full Bayesian multilevel approach for modeling interaction effects in single-vehicle crashes

Area of Science:

• Traffic Safety
• Statistical Modeling
• Transportation Engineering

Background:

• Single-vehicle (SV) crash severity modeling often overlooks crucial interaction effects.
• Spatiotemporal interactions and factor interactions are complex attributes rarely addressed systematically.

Purpose of the Study:

• To develop and validate a Bayesian spatiotemporal interaction multilevel logit (STIML-logit) approach with heterogeneity in means and variances (HMV).
• To systematically uncover interaction effects in SV crash severity modeling.
• To investigate the influence of traffic environment and individual crash factors on injury severity.

Main Methods:

• A full Bayesian STIML-logit approach with heterogeneity in means and variances (HMV) was designed.
• A nested Gaussian conditional autoregressive (CAR) structure was proposed for spatiotemporal interactions.
• Regression modeling was performed on SV crash data from 96 urban roads in Shandong, China, comparing different model specifications.

Main Results:

• The STIML-logit model with HMV demonstrated superior regression performance compared to standard and spatiotemporal models.
• Crash models incorporating a nested CAR structure outperformed those with a traditional CAR structure.
• Significant cross-level factor interactions were identified, indicating that environmental factors' impact on crash injuries varies with case-specific factors.

Conclusions:

• Systematically addressing interaction effects and heterogeneity in means and variances is crucial for accurate SV crash severity modeling.
• The proposed nested CAR structure offers advantages for modeling spatiotemporal interactions in crash data.
• The findings highlight the dynamic influence of traffic environment factors on crash injury severity, dependent on individual crash circumstances.