Analysis of univariate selection index using different statistical methods

The selection index is a linear weighted combination of observed measurements constructed so as to maximize genetic gain. In order to construct an index, estimates of genetic parameters are required. There has been increasing awareness that different approaches for estimating genetic parameters based on likelihood methods provide a suitable framework for statistical inference from animal breeding data. Recent developments in numerical procedures need to be applied to solve practical statistical problems in animal selection for breeding. The thesis is therefore focused mainly on the estimation of variance components and functions of them, and the construction of optimum selection methods using analysis of variance, maximum likelihood and restricted maximum likelihood methods in a balanced univariate half-sib sire model. Data sets used in the thesis are simulated employing the Monte Carlo simulation method. Information on candidates themselves and their relatives are incorporated into a selection index. The conventional theory of selection index using analysis of variance method is compared with the likelihood based ones. Both ANOVA and likelihood based methods, especially REML, give similar results of selection responses due to setting the heritability estimates from ANOVA to zero if it is negative. It is found that the predicted response is more sensitive to the heritability and phenotypic variance estimates than the achieved response. Individual selection and selection index are also compared and it is shown that selection index give more gain than the individual selection. KEY WORDS: Selection index, genetic parameters, likelihood based methods, optimal selection response, achieved selection response, predicted selection response.