Application of Poisson random effect models for highway network screening.

Area of Science:

• Traffic Safety Research
• Statistical Modeling
• Transportation Engineering

Background:

• Traditional methods for crash risk hotspot identification often fail to account for temporal and spatial correlations in crash data.
• Bayesian random effect models are increasingly recognized for their ability to handle complex data structures in traffic safety.
• Accurate identification of crash hotspots is crucial for effective resource allocation and safety improvement initiatives.

Purpose of the Study:

• To evaluate the performance of Bayesian random effect Poisson Log-Normal models in identifying crash risk hotspots.
• To compare the effectiveness of models incorporating temporal and spatial correlations against traditional Empirical Bayesian (EB) and conventional Bayesian Poisson Log-Normal (PLN) models.
• To introduce and validate new methods for assessing the consistency and accuracy of hotspot identification models.

Main Methods:

• Employed random effect Poisson Log-Normal models, specifically considering temporal and spatial correlations (PTSRE model).
• Utilized fatal and injury crash data from urban 4-lane divided arterials in Central Florida (2006-2009).
• Conducted rigorous method examination tests, including site/method consistency, rank difference, total score, and a novel safety performance measure difference test.

Main Results:

• The Bayesian Poisson model with temporal and spatial random effects (PTSRE) demonstrated superior data fitting compared to models with only temporal effects, the conventional PLN model, and the EB method.
• Method evaluation tests confirmed the PTSRE model's significant superiority over PLN and EB models in consistently identifying crash hotspots across different time periods.
• Potential for Safety Improvement (PSI) was used as the measure of crash risk.

Conclusions:

• The PTSRE model, which accounts for both temporal and spatial crash data correlations, is a more effective tool for crash risk hotspot identification.
• This advanced Bayesian approach offers a superior alternative to traditional EB and conventional PLN models for enhancing road safety.
• Consistent and accurate hotspot identification is vital for targeted safety interventions and improved road network safety.