DEFINING THE BEST COVARINACE STRUCTURE FOR SEQUENTIAL VARIATION ON LIVE WEIGHTS OF ANATOLIAN MERINOS MALE LAMBS

In a repeated measures design with two factors, between-subjects and within-subjects, the most appropriate (univariate or multivariate) method and the best covariance structure explaining sequential variation in live weights of Anatolia Merinos lambs fed with different rations were estimated. In general linear mixed model, univariate ANOVA, Geisser-Greenhouse and Huynth-Feldt epsilon were used as univariate methods while profile analysis as well as mixed model methodology were applied as multivariate methods. The data were composed of twenty-four Anatolia Merinos male lambs with weaning age of 2-2.5 months randomly selected from Polatli State Farm and divided equally into four groups. Rations were mas hor pelletted using molasses, lignobond and aquacup binders. Live weights were measured at six times during experimental period (day 0, 14, 28, 42, 56 and 70). In general linear mixed model, nine covariance structures (CS, CSH, UN, HF, AR(1), ARH(1), ANTE(1), TOEP and TOEPH) were applied. AIC, AICC and SBC criteria were used to detect the best defining covariance structure for fitting data. The best covariance structure for the data set was found to be Unstructured (UN). In the case of violation of spherity assumption, using mixed model approach was advised according to AIC, AICC and SBC criteria. As a conclusion, in repeated measures design use of mixed model methodology was recommended to determine the best covariance structure for defining experimental data sets.