A Statistical Regression Analysis Of Domestic Road Accident Influence Factors

Last several decades have witnessed a substantial growth in motorization level,leading a road traffic safety status that has been severe for a long time.However,meanwhile,a large amount of crash data have been recorded simultaneously,which is an invaluable resources for traffic safety research.Analyzing traffic data with regression models,at the macro level,may present the development trend of traffic safety;at the intermediate level,may find potential road influence factor,evaluating crash frequency in a more reasonable way;and at micro level,may estimate the effect of all kinds of driving behaviors on crash severity or crash type.Thus,it is becoming a more and more important auxiliary tool to improve traffic safety performance.Firstly,this study proposes a three-level framework concept about road traffic crash data grouping and its research methodology.Three works of each level,the macro,the intermediate and the micro level,are done,with domestic road traffic crash data in different ranges.At the macro level,considering the tremendous increasement in ownership of electric bicycle per capita and a long-term time trend of the improvement of traffic safety consciousness,based on Smeed model and an improved form,two modified Smeed models are established,using recent years' traffic accident death data collected from the Ministry of health.The results show that the short-term domestic macro level road safety performance will maintain upgrade in a gradually slow-down pattern.At the intermediate level,with a whole 3-years' crash date set gained from a road of a city in China's eastern region,taking data properties and urban road particularity into account,for researching monthly crash frequency of each segment,a hierarchical mixed effect joint negative-binomial model is estimated by MCMC full Bayesian method.The results show that this model is superior to standard Poisson model and Negative-binomial model in the terms of goodness-of-fit;the land use pattern surrounding the road may influence the crash frequency;the presence of “Beng-Beng”,a three-wheeled passenger motorcycle,is potentially a road traffic safety hazard;in addition,independent variable time,signal control,number of lanes,side barrier type,density of pedestrian crossing,and density of horizontal curve are also significant.At the micro level,with a 5-year general crash data set gained from a region inwestern China,concentrating on the injury severity of drivers,a hierarchical mixed effect multinomial logit model and a hierarchical mixed effect ordered logit model are estimated by MCMC full Bayesian method.The results show that the variable “diff”adequately interprets the diffierence of types between the two objects in a crash;part of the illegal behaviors have significant effects on drivers' injury severities;when collided by a speeding vehicle,or driving a vehicle without plate while crashing,the driver may suffer more serious injury.Finally,integrating conclusions drawn from the researches at the above three levels,general recommends for better regression analysis of crash data are suggested,and remarks directing further researches of each level are discussed.