A Bayesian spatial Poisson-lognormal model to examine pedestrian crash severity at signalized intersections

Area of Science:

• Transportation Engineering
• Traffic Safety Research
• Urban Planning

Background:

• Nonmotorized crash reduction necessitates understanding causes and consequences at the facility level.
• Existing crash models lack crucial pedestrian exposure data and fail to capture correlations across crash aspects.

Purpose of the Study:

• Develop a robust framework for pedestrian crash analysis using a multivariate model.
• Incorporate pedestrian exposure data to examine multiple crash severities simultaneously.

Main Methods:

• Employed a multivariate spatial (conditional autoregressive) Poisson-lognormal model within a Bayesian framework.
• Analyzed data from 409 signalized intersections in the Austin area.
• Included traffic characteristics, road geometry, built environment features, and estimated pedestrian exposure volume.

Main Results:

• The multivariate model outperformed univariate models, confirming the need to jointly analyze multiple crash severities.
• Higher speed limits positively influenced fatal pedestrian crashes.
• Increased motorized traffic volume correlated with higher incapacitating and non-incapacitating injury crashes.
• Bus stop presence had mixed effects: negative on incapacitating, positive on non-incapacitating injury crashes.
• Higher pedestrian volume at intersections positively influenced non-incapacitating injury crashes.

Conclusions:

• Jointly modeling multiple pedestrian crash severities is essential for accurate analysis.
• Factors like speed limits, traffic volume, and pedestrian volume significantly influence crash outcomes.
• Tailored intersection design and policy are crucial for mitigating pedestrian crashes across all severity levels.