| The count data are widespread in agriculture, medical, public health, finance, insurance, and many other areas. The basic model of analysis these data is the classic discrete regression model, such as Poisson regression model, negative binomial regression model and so on. Some of count data can appear a lot of zero data, such as the number of pneumonia in the farm animals, the number of eggs in animal excrement and so on. The zero data significantly more than the Poisson distribution, negative binomial distribution and other discrete distribution produces zero, we call this kind of data is zero inflation data, which is zero too much data. Zero inflation data has been more and more attention in various fields. In recent years, scholars at home and abroad put forward many model to deal with zero inflation data under the various background. For example the ZIP model, ZINB model, Hurdle model and so on. In this paper, the main research object is this kind of special data, the typical model is introduced in detail, and applied in the field of veterinary epidemic. Concrete research content is as follows.This paper first describes the research background. The second chapter introduces the Poisson regression model and the negative binomial regression model, and the Poisson regression model is applied in the field of veterinary epidemic to do an instance analysis, but the result is bad. The third chapter introduces the parameter estimation method of ZIP model and ZINB model, parameter estimation and Score test are simulated on the case of ZIP model, Compare advantages and disadvantages under different zero ratio, the result is ZINB model estimation precision is higher than other models for zero data in the case of excessive discrete, then ZINB model is used to analyze the waste eggs data in the field of veterinary epidemic, calculate the Score test statistics is77.309, explain that ZINB model is more suitable for the data than the negative binomial regression model. In the fourth chapter introduces the Hurdle model, include parameter estimation method, simulation study and data deleted model statistical diagnosis, and use Hurdle-NB model to analyze waste eggs data in the field of veterinary epidemic, Statistical diagnosis find that978points have strong impact. The fifth chapter gives the research result in this paper. Result showed that the Poisson regression model, the negative binomial regression model is the basis of the simulation count data model, Poisson regression model requires data expectation and variance are equal, the negative binomial regression model is suitable for a certain degree of discrete count data, but when the data contains large number of zero, the above two models is not ideal, and zero inflation model and Hurdle model are more suitable for zero inflation data. From the instance analysis found that in the field of veterinary epidemic ZINB and Hurdle-NB model have good effect for zero inflation data. Hurdle model also can analyze the zero shrinkage count data, that is zero less data. Specific application needs further study. |
