Use of generalized estimating equations for lactation milk yield

Generalized Estimating Equations (GTD) are used effectively to produce regression estimates in data sets containing repeated measurements, panel data, grouped data, and non-normally distributed data. The results of the variance that has changed between longitudinal measurements taken from the same unit in repeated data are significant in terms of statistical analysis to determine the variance-covariance structure. The structure belonging to variance-covariance is directly effective in determining the working correlation structure. This method was developed for discrete and continuous longitudinal data. However, it can also be used on related data. This thesis aims to determine the most appropriate correlation structure between repeated measurements that make up the data set and the most appropriate method, using the GTD. This thesis study was conducted at Ondokuz Mayıs University, Faculty of Agriculture, Department of Animal Science. In the study, lactation duration, 305-day milk yield, total lactation yield, and lactation duration parameters of 228 head of animal from 3 cattle breeds belonging to 2 enterprises were used. In addition, correlation structures created with GTD (AR(1), Interchangeable, Independent, M-Dependent, and Unstructured), estimates (Maximum Likelihood, Deviance, and Pearson Chi-Square), and (QIC, QICC) information criteria were used. For each correlation structure, the structure with the lowest information criteria was selected as the model. Following this process, the chosen correlation structure represents the best model. Data from 228 head of dairy cattle were subjected to statistical analysis. According to the analysis results made with the GTD method, it was determined that the structure of each correlation model was different from each other. In addition, after applying 3 different estimation methods, the QIC value of the selected model was calculated. As a result, the "Independent Correlation Structure" was determined to be the best model representing the selected model.