Application of generalized estimating equations (GEE) for real longitudinal data

Every statistician's goal is to obtain optimal statistical models and then estimate the parameters of these models to make inferences. Selecting an appropriate statistical method actually requires careful thinking about how data should be collected and what they measure, but normally raw data is sometimes overcome the classical assumptions of statistics which inhibit access to good estimates. We addressed the problem of correlated data which is one of the most important assumptions in estimation methods. Many solutions were developed to solve this problem; however our goal is always to get sufficient solution. The Generalized Estimating Equations (GEE) have become a popular approach that provides a sufficient solution for the data that occur in correlation condition. An important feature of the GEE methodology takes into account the type of correlation effect ("working correlation structure"). In this thesis, initially we will give an overview of Generalized Linear Models (GLM) algorithms and their techniques of getting estimates; which the GEEs had presented in the GLM environment. In addition GLMs have an important utility that is ability to derive statistics and properties for group of data which based on likelihood, that makes statisticians create methods within the GLMs framework, but it require independence where sometimes is not applicable. Additionally, we will make comparison for estimated parameter and standard errors between models which solved the correlated data and GEE. There are many kinds of correlated data in social sciences; we used longitudinal data as an example. Key Words: Generalized Estimating Equations (GEE), Quasi Likelihood, Longitudinal Data, Correlation Effect