Analysis Of Factors Affecting Traffic Accident Severity Based On Heteroskedasticity Ordinal Logit

With the development of society and economy,road infrastructure has been continuously improved,and the number of motor vehicles has rapidly increased.By 2016,the number of private cars in China had reached 165.59 million.However,the traffic accidents that followed were frequent and 34,774 people died in traffic accidents in 2016,causing huge losses to the country.In order to accurately grasp the law of accidents and formulate scientific accident prevention and control measures to reduce the number of casualties,the existing research has introduced statistical models and econometric theories to analyze a great deal of accident data.However,due to the diversity of accident data and limitations of analytical models.there are still many problems in the current accident analysis.To clarify the applicability of the method and improve the analysis accuracy.First,on the basis of drawing lessons from domestic and foreign research results,this study summarized the severity categories of road traffic accidents,analyzed and summarized the mechanism of accident evolution,and classified the factors affecting the severity of accidents according to five categories: person,vehicle,road,environment,and management.and introduced a covariance theory to analyze the interaction among factors.According to the accident characteristics,an applicability analysis of discrete selection model was conducted.Modeling,estimating and testing theories of fixed variance logit models,including ordinal logit model and multiple logit model,were summarized for studying the accident severity.Then a complete research process is constructed.After that,in order to test the variance characteristics of crash data,for the first time,two econometric diagram methods,the White test method and the Braut-Pegan-Godfrey test method,were introduced.Through the introduction of heteroscedasticity theory,an accident data analyzing method based on heteroscedasticity ordered logit model was formed.Finally,the method was verified using crash fixture accident data.The research results indicate that road traffic accident data has heteroskedasticity,the fixedvariance logit model has better applicability to independent and identically distributed data,and heteroskedasticity ordinal logit model is more suitable for accident data with variable variance because of the lesser restrictive hypothesis,so the model can effectively capture factor heterogeneity and tap into more potential latent variables.the most significant factors that influence the crash fixture accident severity are driver safety measures(p<0.0000)and chemical tests(p<0.0000),which have 99% confidence level in three regression models.The vehicle type factor(p<0.0590)is also statistically significant with a 94% confidence level,heteroskedasticity ordinal logit model found that the right shoulder type had a significant impact on the severity of the accident with a confidence level of 99%.