Modelling the severity of pedestrian injury in pedestrian—vehicle crashes in North Carolina: A partial proportional odds logit model approach

Abstract

More than 2,000 pedestrians are involved in reported traffic crashes with vehicles in North Carolina every year. Of these, 10% to 20% of these pedestrians were killed or severely injured. Studies are needed to explore the reasons under such situation for improvement. Traditional methods for modeling crash injury severities use multinomial logit (MNL) models, mixed logit (ML) models, or ordered logit/probit models. However, traditional MNL and ML models treat injury severity levels as nonordered, ignoring the inherent hierarchical nature of crash injury severities, whereas the data used in ordered logit models should be strictly subjected to the proportional odds (PO) assumption. A partial proportional odds (PPO) logit model approach is applied here to address these concerns. The predictors in the PPO model can have different effects on different levels of the dependent variable who violates the PO assumption. This study uses police-reported pedestrian crash data collected from 2007 to 2014 in North Carolina. A variety of motorist, pedestrian, environmental, roadway characteristics are examined. Comparisons of different models are also made and results show that PPO outperforms others. Parameter estimates and associated marginal effects are calculated and used to interpret the model and evaluate the significance of each dependent variables, followed by recommendations in the Conclusion.