Genel Doğrusal Karışık Modellerde Farklı Kovaryans Yapıları ve Tahmin Yöntemlerinin Performanslarının Karşılaştırılması

In studies on analysis and interpretation of the data with repeated measures structure, an enormous progressive has been showed in recent years and in this sense, the very strong methods have been developed. Three models which the time variable is included in different form to model have been established, benefitting the specific situations of linear mixed model. This models constituted as the random intercept and slope model which the time was included in continuous variable (Model 1), the random intercept model which the time was included in categorical variable (Model 2) and the random intercept and slope model which time was included in both continuous and categorical variable (Model 3). Compound Symmetry (CS), Unstructured (UN) and First Order Autoregressive (AR(1) structure were applied in determination of the covariance structure between repeated measures, and along with these structure, Maximum Likelihood (ML), Restricted Maximum Likelihood (REML) and MinimumVariance Quadratic Unbiased Estimator (MIVQUE) were used estimation methods. Selection of the best adequate estimation method and covariance structure for dataset was evaluated by AIC and BIC criteria. Data were taken from values of serum testosterone concentration, which it was collected at 33 of Norduz male lambs. In conclusion, it was revealed that the best cohesion with dataset was shown by the UN covariance structure taking into account a heterogeneous structure along with ML estimation method in all three models