Determination of Best Variance-Covariance Structure in Mixed Model (SAS Proc Mixed) with Various Parameter Estimation Methods.

The aim of this study was to compare the covariance structures by using Maximum Likelihood (ML), Restricted Maximum Likelihood (REML) and Minimum Variance Quadratic Unbiased Estimator (MIVQUE) in the estimation methods in repeated measures design with mixed model approach. In the study, live weight (birth, 30th, 60th, 90th, 120th day) values of 60 head Kilis goats from birth to 120 days old were used as research data. For the purpose of evaluate of the relationship among the data, Compound symmetry (CS), Variance components (VC), (First-order autoregressive (AR(1)), Unstructured (UN), Toeplitz (TOEP), Heterogenous compound symetry (CSH), Heterogenous first-order autoregressive (ARH(1)), Heterogenous toeplitz (TOEPH), First-Order Autoregressive Moving-Avarege (ARMA(1,1)), Toeplitz With Two Bands (TOEP(2)), First-order factor analytic (FA(1)), Equal Diagonal Factor Analytic (FA1(1)), Unstructured correlations (UNR), Banded Unstructured (UN(1)), Ante-Depence (ANTE(1)) covariance structures were used. The most appropriate covariance structure was selected according to 2Ln(L), AIC, AICC and BIC information criteria in modeling the relationship between data in all three estimation methods (ML, REML and MIVQUE0), UN and UNR covariance structures were determined as the most appropriate covariance structures, although they gave the same results. [ABSTRACT FROM AUTHOR]