Statistical Analysis Methods Of Crash Risk And Injury Severity Of Pedestrian-Vehicle Crashes At Intersections

Walking is the most basic mode of transportation,which is essential in almost all travel processes.Walking safety is closely related to each road user.However,in many countries and regions,pedestrian traffic safety is still a serious problem.According to the World Health Organization,pedestrians account for about 23% of the deaths from road traffic crashes.Laws of the crash can be found in the crash data.Using statistical methods to analyze pedestrian-vehicle crash data at intersections,exploring the impact factors on the occurrence of crashes and injury severity,can provide reference for the actual traffic safety work,which is of great significance to reduce the occurrence of pedestrian-vehicle crashes at intersections and reduce injury severity.This study focuses on exploring the law of pedestrian-vehicle crashes at intersections,analyzing the crash risk of pedestrian and driver violations,and studying the influencing factors of pedestrian injury severity.Based on the data of pedestrian-vehicle crashes at intersections in the historical traffic crash database,this study uses statistical methods to analyze pedestrian,driver,vehicle,road and environment related factors.First,the basic distribution characteristics of pedestrian-vehicle crashes were obtained.Second,the crash risk analysis of pedestrian-vehicle crashes at intersections was carried out based on the quasi-induced exposure technique.The crash risk of pedestrian violation and driver violation under different factors was calculated and analyzed.Besides,the significant influencing factors of pedestrian violation and driver violation were identified using binary Logistic regression model.Considering that different traffic control devices at intersections may have different impacts on people’s behavior,this study analyzed three types of intersections separately: no control,signal control,and stop or yield sign control.Some of the main conclusions are as follows: male pedestrians,pedestrians under 18 years old,pedestrian under the influence of alcohol or drug,and unmarked crosswalk would increase the crash risk of pedestrian violation behavior at all the three types of intersections;at uncontrolled intersections,the crash risk of pedestrian violation behavior at night is significantly higher than that in daytime;the crash risk of left turn and right turn vehicles is higher than that of straight vehicles at all the three kinds of intersections;the crash risk of driver violation behavior at signalized intersections would significantly increase under the situations of speeding,adverse weather,and at the farside of intersection,etc.Finally,the injury severity analysis was conducted based on the latent class clustering analysis.Considering the data heterogeneity between groups with different characteristics and the unobserved heterogeneity within each group,the latent class clustering method and the random parameter ordered Probit model were comprehensively applied,and the models’ goodness-of-fit were compared.The relationships between significant factors and pedestrian injury severity of each cluster were analyzed and discussed.The results indicate that the random parameter ordered Probit model based on latent class clustering could better capture the heterogeneity of data,and could obtain more specific factors,though it may also increase the complexity of the model.In terms of influencing factors,urban area,pedestrian age of 41-65 years old and over 65 years old,pedestrian under the influence of alcohol or drug,farside of the intersection and pedestrian violation,etc.,have significant impacts on all clusters,which indicates that these factors are highly correlated with pedestrian injury severity.While some factors were found only significant in the cluster-based models,such as 6am-10 am,dawn/dusk,male pedestrian,driver age of 41-60 years old and over 60 years old,which reveals that these factors have significant impact on crashes with specific characteristics and these potential important factors could not be identified if only the whole data was analyzed.