Hurdle-generalized Poisson Regression Model And Its Statistical Diagnosis

Count data is a common kind of discrete data,there are a lot of count data in our daily life involved in many fields. The basic model of analysis these data is Poisson regression model,but the application of such models implies that the sample variance should be equal to the mean,in practice, however, this is hard to satisfy.The gener-alized Poisson model is a extension of the standard Poisson model ,it introduces a dispersion parameter which can weight the relationship between the mean and vari-ance.On the other hand,some of count data can appear a lot of zero data,The zero data significantly more than the Poisson distribution,the generalized Poisson model produces zero,we call this kind of data is zero too much data.In this paper,the main research object is the kind of special data which concludes zero too much and the sam-ple variance is not equal to the mean,the typical model-Hurdle-generalized Poisson regression model is introduced in detail.Concrete research content is as follows.This paper first describes the research background. The second chapter intro-duces the generalized Poisson regression model and the Hurdle regression model,and the Hurdle-generalized Poisson regression model,at the same time,it gives the pa-rameter estimation method of the Hurdle-generalized Poisson regression model.The third chapter introduces the data deleted model statistical diagnosis,and gives the one step approximations of the estimates,generalized Cook distance and likelihood distance,And gives the test of the existence of the dispersion parameters.The fourth chapter introduces a method of selecting a model,Through these guidelines to deter-mine which model fitting effect is better. The fifth chapter uses the stochastic simu-lations to test effectiveness of the statistics which introduced in the second and third chapter.The sixth chapter through the example of ear disease occurs to illustrate it has a better fitting effect using the Hurdle-generalized Poisson regression model for the data which contained too much zero and expectation is not equal to variance of data. Paper finally gives the conclusion and some issues worthy of further study.