The Zero-Inflated(ZI)model has been used in many fields such as insurance actuarial,health care,and socio-economic.It is used to explain the problem of excessive or excessive dispersion of count data,such as the zero-inflation negative binomial model can handle over-dispersed data,and the zero-inflation generalized poisson model can also Handle count data with too many zeros or over-dispersion.However,the traditional zero-inflation regression model assumes that count data is uncorrelated.In this case,the existing zero-inflation model may not be able to accurately analyze the characteristics of count data.Therefore,the Generalized Estimation Equation(GEE)is used to solve this problem,which is used to estimate the parameters,and at the same time can explain the correlation between observations from the same subject,Therefore,this paper proposes a generalized estimation equation(GEE)based on the zero-inflated generalized Poisson(ZIGP)regression model to explain the excessive dispersion of count data,excessive zeros,and the correlation between observations from the same subject.In this study,three different methods were used to analyze the data in the Iowa fluoride research,respectively,using the generalized estimation equation estimation method based on the zero-inflated generalized Poisson regression model,the independent ZIGP model and the GEE method without the zero-inflated model.the data were fitted,the response variable was the number of caries of the subject,the covariate was the age in years at the time of the dental examination,the fluoride intake by the subject from the diet,the proportion of the number of dentist visits 6 months before each observation time point of the subject,the average number of times the subject received professional fluoride treatment for the tooth 6 months before each observation time point and the average daily brushing frequency of the subjects.By fitting these three different methods and conducting a comparative analysis,Indicating that the generalized estimation equation estimation method based on the zero-inflated generalized Poisson regression model fits better and is better than the other two methods. |