Related Experiment Videos
Modeling accident frequencies as zero-altered probability processes: an empirical inquiry
V Shankar1, J Milton, F Mannering
1Department of Civil Engineering, University of Washington, Seattle 98195, USA.
Accident; Analysis and Prevention
|November 25, 1997
Summary
Zero-altered counting processes effectively distinguish safe roadways from those with unobserved accidents. These models improve accident prediction by accounting for zero-accident observations, aiding highway safety design.
Area of Science:
- Transportation Engineering
- Statistical Modeling
- Road Safety Analysis
Background:
- Traditional accident frequency models (Poisson, negative binomial) often yield biased estimates due to numerous zero-accident observations.
- These models fail to differentiate between roadways with genuinely low accident risk and those with zero observed accidents by chance.
- This limitation hinders accurate identification of factors contributing to road safety.
Purpose of the Study:
- To investigate the application of zero-altered counting processes for analyzing roadway section accident frequencies.
- To differentiate between roadways with true near-zero accident likelihood and those with observed zero accidents.
- To enhance the accuracy of accident prediction models for improved highway safety management.
Main Methods:
- Empirical inquiry into zero-altered counting processes, specifically zero-inflated Poisson (ZIP) and zero-inflated negative binomial (ZINB) distributions.
- Application of these models to accident frequencies categorized by roadway functional class and geographic location.
- Utilizing the Vuong specification test for non-nested hypotheses to compare model performance.
Main Results:
- Zero-inflated Poisson (ZIP) models demonstrate promise and flexibility in analyzing accident frequencies.
- These models successfully uncover processes influencing accident occurrences on roadways with both zero and non-zero observed accidents.
- The findings highlight the ability of ZIP models to better isolate design factors affecting accident rates.
Conclusions:
- Zero-altered models offer a superior approach to analyzing roadway accident data, particularly with excess zero counts.
- Highway engineers can leverage these flexible models to better understand accident causation and identify safety improvement opportunities.
- The proposed models provide a foundation for a comprehensive family of accident prediction tools within safety management systems.