The Comparative Study Of Hurdle And Zero-Inflated Regression Model In Traffic Accident Casualties

Zero expansion count data exists widely in areas such as agriculture, medical, industrial, insurance and so on and Hurdle regression and zero inflation regression models are the two types of most effective regression models to analysis and deal with these data. Hurdle and zero-inflated regression models are proposed for count data with zero expansion characteristics. ZIP and PH regression model is commonly used for dealing with zero expansion count data, but both can only solve the problems of zero expansion.ZINB and NBH regression model as a expansion of ZIP and PH regression model, can not only solve the problems of zero expansion, but take the over-dispersion from the unobserved heterogeneity and correlation into consideration. With the rapid development of economy, traffic accidents happen more frequently, causing wide attention of scholars both at home and abroad. Traffic accident casualties is a count data with zero-inflation characteristics, which can be fitted with Hurdle regression and zero-inflated regression model. At the same time, application research of the two models is generalized to the new field of traffic accidents. As PH and ZIP regression models fit the data of traffic accident data, the zero inflation is considered, which can improve the general counting regression model effectively and make the conclusion more accurate. NBH and ZINB regression models, as expansions of the PH and ZIP regression models, are the optimal theoretical models for the data of casualties in traffic accident which has the characteristics of not only zero expansion but over-dispersion caused by individual heterogeneity between discrete characteristics.In this paper, different counting regression models are applied to analysis the influence factors of traffic accident casualties based on the research on traffic accident data, and compared by the model selection methods, drawing the following conclusions:1.More traffic accidents happen in the intensive month and time of travel. Significantly more accidents happen in sunny days than bad days. Most of the traffic accidents cause property damage of 1000 Yuan approximately and the relatively few accidents lose more than ten thousand Yuan. The drivers of Accident vehicles drivers are mostly between 23 to 45 years old.2. ZINB regression model is the best suitable model in analysis of factors influencing traffic accident for the data.3. The license, damage, weather conditions and drivers’qualifications of the influencing factors of traffic accident have a significantly influence on the number of casualties.4.The conclusion of ZINB regression model in the analysis of the influence factors of traffic accident casualties in the analysis results is consistent with that of the cross analysis, further verifying that the ZINB regression model is applied in the analysis of factors affecting traffic accident casualties in rationality.This research only studies the affecting factors on the number of casualties like the weather, the vehicle models, license, vehicle damage, the driver’s age, driving experience and so on, but Hurdle and zero-inflated regression models are more suitable in large amount of data. In order to get better results, a large number of data can be collected and the more influencing factors can be selected.