Comparative Study On Rollover Crash Injury Severity Estimation Based On Heterogeneous Effects Models

In road traffic crashes,although rollover crashes account for a relatively low proportion,result in a high fatality rate.Therefore,domestic and foreign scholars have made certain achievements in the research on rollover crashes,but there is still some room to study the rollover crash injury severity.Most of the existing crash data are unable to observe or record all the information related to the rollover crash,such as the driver's height and weight,and the existence of these unobservable or recorded rollover crash information may lead to the different impact of the same safety factor on rollover crash injury severity in the statistical analysis of risk factors associated with crash,thereby producing different effects,which is also known as heterogeneous effects.However,the existing research has less considered the heterogeneous effects of various safety factors on the rollover crash injury severity.Ignoring the heterogeneity that may exist in the crash data may result in a biased estimation dof the model coefficients.In addition,most of the existing studies did not use different models for comparative analysis of the rollover crash injury severity,leading to unreliable results.Firstly,this paper expounds the concept of heterogeneity in rollover crash data and analyzes the heterogeneity sources of the influencing factors of rollover crash injury severity by combing the definition of heterogeneity in different fields.On the basis of borrowing and absorbing domestic and foreign achievements,this paper summarizes the modeling,solving and testing theory of multinomial logit model without considering heterogeneous effects and the mixed logit and latent logit heterogeneous models.A complete analysis process for crash injury severity analysis was developed.Secondly,based on the rollover crash data in Texas in 2016,this paper selected 20 safety factors as independent variables from four aspects of people,vehicles,roads and environment and considered the rollover crash injury severity as the dependent variable.The mixed logit and latent logit heterogeneous models and the multinomial logit model without considering heterogeneity effects were developed,and the goodness of fit and prediction accuracy of the three models are compared.Thirdly,the results of the mixed logit model and the latent logit model are compared from the heterogeneous results and the prediction probability.Finally,by calculating the average marginal effect of each significant variable in the latent logit model,the quantitative impacts of each significant safety factor on the rollover crash injury severity is analyzed,and the targeted rollover crash prevention countermeasures are proposed according to the analysis results.The three model comparison results above show that the rollover crash injury severity models considering the heterogeneous effects are better than that without considering the heterogeneous effects,indicating that the mixed logit and latent class logit models can solve the heterogeneity problem in traffic crash data to a certain extent.In addition,the latent class logit model is superior to the mixed logit model in terms of the ability to capture heterogeneous effects,goodness of fit and prediction accuracy.The the estimation results clearly indicate the similarities and differences between mixed logit and latent class logit models.A total of 16 safety factors(such as driver's gender,driver's age)were found to be significant in both models,while two factors such as vehicle driving intention,and the road alignment were found to be significant only in the mixed logit model.Alternatively,vehicle types,speed limits of the road section,seat belt use or not,and driver's age were found to have heterogeneous effects on rollover crash injury severity in the latent class logit model,while driver's gender,driver's age were also found to be heterogeneous factors associated with rollover crash injury severity in the mixed logit model.These research findings can not only explore the heterogeneous effect mechanism of various influencing factors on rollover crash injury severity in-depth,but also can provide a theoretical basis for proposing targeted preventive measures for rollover crashes.In addition,this paper can provide a paradigm for follow-up domestic analysis of crash injury severity.