Small Sample Correlated Binary Data Analysis Based On GEE

Correlated binary data has been widely used in various fields and has aroused greatconcern of all researchers.For this type of data,generally,the following statistical methods can be used to estimate: full likelihood(FL)method,penalized quasi-likelihood(PQL)method,generalized estimating equations(GEE)method,fixed-effects logistic regression(FELR)method.In this paper,the generalized estimation equations(GEE)that takes into account the correlation between observations in a cluster is used for the study,the numerical simulation of the small sample data is used to verify the fitting effect of the small sample data,find the factors that affect the model fitting effect,we used this method in real data analysis.This paper mainly includes four parts: The first part introduces the background of this topic,as well as the existing work analysis of the generalized estimation equations by domestic and foreign scholars.The second part analyzes the Logistic regression and generalized linear model(GLM),then introduces the knowledge of generalized estimation equations and the correction of the bias of the estimation of generalized estimation equations;the third part discusses the influence of the sample size,the correlation coefficient and the error-specified structure on the model fitting effect through the simulation study.And observe the efficiency of the deviation correction of the regression parameters.The fourth part is an example analysis of the clinical research on the treatment effect of patients with respiratory diseases.