A Comparative Study Of Zero Expansion Models And Test Methods

Count data is an important class of data,which always appear in medicine,biological and agricultural areas.When the proportion of zeros in the count data exceeds the predictive capacity of the common count model such as the Poisson regression model or the negative binomial regression model,that is,there is Zero-Inflated phenomenon in the count data,it is possible to give misleading statistical inference by fitting the data with the common count model such as the Poisson regression model.Therefore,investigation the existence of Zero-Inflated phenomena in the count data for selecting an appropriate model is statistically significant.This research analyzes these two issues based on the Monte Carlo simulation,the specific contents are as followsIn the first chapter,we presented a brief introduction,literature review as well as the concept and main contents of the research.In the second chapter,we introduced the Poisson regression model,NB regression model,ZIP regression model and ZINB regression model and the concern knowledge.In the third chapter,using the Monte Carlo simulation method,test whether the Zero-Inflated phenomenon exist in the count data under different levels of Zero-Inflation,different of mean and sample size,examine the efficiency of the seven test methods and the probability of the type I error under the Zero-Inflated Poisson regression model for elucidating the properties of the seven test methods.In the fourth chapter,the Monte Carlo simulation method have be used to generate the random data which are ZIP distribution and ZINB distribution and fitted the generated random data by using Poisson regression model,NB regression model,ZIP regression model and ZINB regression model to investigate the advantage of the models under different of model selection criteria.In the fifth chapter,the results of this study have be summarized and some of shortcomings for future study have be given.