Modelling severity of pedestrian-injury in pedestrian-vehicle crashes with latent class clustering and partial proportional odds model: A case study of North Carolina

•Pedestrian injury severities are modeled in pedestrian-vehicle crashes in North Carolina.•Latent class clustering approach is applied to segment the crash data for reducing the heterogeneity in data.•Partial proportional odds models are developed to further explore the contributing factors to pedestrian injury severities.•Differences in injury-severity probabilities are found between each latent class.•Key contributing factors are identified, and associated policy recommendations and future research directions are given. There are more than 2000 pedestrians reported to be involved in traffic crashes with vehicles in North Carolina every year. 10%–20% of them are killed or severely injured. Research studies need to be conducted in order to identify the contributing factors and develop countermeasures to improve safety for pedestrians. However, due to the heterogeneity inherent in crash data, which arises from unobservable factors that are not reported by law enforcement agencies and/or cannot be collected from state crash records, it is not easy to identify and evaluate factors that affect the injury severity of pedestrians in such crashes. By taking advantage of the latent class clustering (LCC), this research firstly applies the LCC approach to identify the latent classes and classify the crashes with different distribution characteristics of contributing factors to the pedestrian-vehicle crashes. By considering the inherent ordered nature of the traffic crash severity data, a partial proportional odds (PPO) model is then developed and utilized to explore the major factors that significantly affect the pedestrian injury severities resulting from pedestrian-vehicle crashes for each latent class previously obtained in the LCC. This study uses police reported pedestrian crash data collected from 2007 to 2014 in North Carolina, containing a variety of features of motorist, pedestrian, environmental, roadway characteristics. Parameter estimates and associated marginal effects are mainly used to interpret the models and evaluate the significance of each independent variable. Lastly, policy recommendations are made and future research directions are also given.