A random parameter negative binomial model for signalized intersection accidents in Seoul, Korea.

Area of Science:

• Traffic Safety Research
• Statistical Modeling
• Transportation Engineering

Background:

• Traditional statistical models for crash analysis often use fixed parameters, limiting their ability to capture time-varying or site-specific effects.
• Existing models like Poisson and Negative Binomial struggle to account for unobserved heterogeneity in accident data.

Purpose of the Study:

• To develop and apply a random parameter negative binomial model for traffic accident frequency analysis.
• To identify factors influencing traffic safety at signalized intersections, considering unobserved heterogeneity.

Main Methods:

• Utilized a four-year (2007-2010) panel dataset of accident histories from 95 signalized intersections in Seoul, Korea.
• Estimated a random parameter negative binomial model incorporating traffic volumes and intersection geometric characteristics.

Main Results:

• Identified the presence of a left-turn exclusive lane, median barrier (existence and length), and pedestrian island on a major road as significant random parameters.
• Ten additional variables, including heavy vehicle volume, exclusive turn lanes, taxiway lanes, median barriers, and number of lanes on major and minor roadways, were significant fixed parameters affecting intersection safety.

Conclusions:

• The random parameter negative binomial model effectively accounts for unobserved heterogeneity in intersection accident data.
• Findings highlight the importance of analyzing lane channelization, lane exclusion, and lane geometry as potential random parameters to improve intersection safety predictions.