Study On Factors Contributing Crash Severity Of Rural Highway Considering Temporal And Spatial Correlation

Traffic crash is the result of the interaction between human activities and different socio-economic,geographical environment and other factors,showing a temporal and spatial relationship.The temporal and spatial correlation needs to be characterized in the study of crash severity.For this,geographically and temporally weighted ordered logistic regression(GTWOLR)model is an effective method.However,the existing researches using GTWOLR model only subjectively selected one form of kernel function and kernel bandwidth,which cannot determine the best expression of the spatial-temporal relationship between crashes.Therefore,the optimal kernel function and kernel bandwidth for crash data need to be further explored.In view of the above problem,the thesis firstly selected 25 independent variables from the data on rural highway crashes(2014-2017)in Anhui Province,China.Secondly,the GTWOLR models,with Gaussian or Bi-square kernel function,fixed or adaptive bandwidth,were constructed respectively,and the goodness of fit of the models were compared to explore the optimal kernel function and kernel bandwidth.Thirdly,the GTWOLR model and ordered logistic regression model were compared by using log likelihood,Pseudo-R2 and Akaike Information Criterion.Finally,based on the analysis of the contributing factors and non-stationary test results of GTWOLR model,the corresponding countermeasures to improve the traffic safety were proposed.The comparative results of GTWOLR models showed that the Bi-square kernel function and fixed bandwidth can best adapt to crash data.Compared with the ordered logistic regression model,GTWOLR model provides better model goodness of fit.The GTWOLR model estimation results showed that eight factors passed the non-stationary test,including pedestrian-vehicle crash,middle-aged driver,hit-and-run,truck,motorcycle,curve,slope and mountainous,indicating that their effects on the crash severity vary across space and over time.The optimal kernel function and kernel bandwidth for crash data determined in the thesis have certain reference significance for future research.In addition,the results in the thesis can help traffic management departments to propose forward-looking and targeted policies or countermeasures,so as to reduce the severity of rural highway crashes more effectively.