ESTIMATION OF THE VARIANCE-COVARIANCE COMPONENTS IN MIXED MODELS - A MONTE-CARLO COMPARISON OF SEVERAL METHODS AND POPULATION SIZES IN LAYING HENS

Four methods of estimating variances and covariances were compared via selection for a mixed model in laying hen breeding with hierarchical structure of the random effects and with unbalanced data. The methods applied are: ANOVA, analysis of variance with correction of the fixed effects (ANOVAKOR), MIVQUE(0) and REML. Sire, granddam, dam and error effects were simulated at random from normal distribution for four traits (egg number, egg weight, body weight, feed conversion) and for two population sizes (N1 approximately 1800, N2 approximately 500). The simulation was replicated 75 and 135 times respectively. The estimators were compared through their biases, MSE's and the achieved responses to selection using multitrait selection indices obtained by the four methods of estimation. Between ANOVA and the other methods there are larger differences in bias, MSE and achieved responses to selection. Distinct advantages for REML are especially upon N2 and also for the variance of selection gain.