Abstract
Objectives
Highly aggregated data is conventionally used in transportation safety omitting seasonal variations and leading to important loss of information. States in the northern region of the United States experience significant weather variations with snowfall and ice events. Crash occurrence is the highest during the winter compared to other seasons. Therefore, seasonal safety performance models would enable better safety analysis and decision-making. A new modeling approach is proposed which takes into account seasonal crash prediction. A case study is provided with crashes, operational, and weather data of urban signalized intersections from the southeast region in Wisconsin. Four seasons were considered: winter, spring, summer, and fall.
Methodology
The modeling approach consisted of the Multivariate Negative Multinomial to account for seasonal variations and severity classification. Functional forms of predictor variables were optimized. Measures of log-likelihood, Overdispersion, Cumulative Residual (CURE) plots, and Akaike Information Criterion (AIC) showed adequate model prediction accuracy.
Results
Crash estimates were the highest during the winter and the lowest during the spring seasons. Estimates remained below or near the annual average for the summer and fall seasons. Model performance was evaluated and results showed that seasonal model prediction and observed crash rate variations as percentage of annual estimates were similar.
Conclusions
Seasonal estimates have a significant contribution in safety analysis in regions where snow and ice conditions are regularly experienced. Identifying locations that experience significantly higher number of crashes during the winter can contribute to target snow and ice-related crashes and evaluate the effectiveness of deicing materials, equipment, and practices.
Acknowledgments
The authors are thankful for the assistance provided by Wisconsin DOT staff Brian Porter, Kevin Scopoline, and Jarrett Gates. The authors wish to acknowledge the contributions of Eric Winkelman and Alec Nelson, who assisted with data collection. The authors received no financial support for the research, authorship, and/or publication of this article. The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.