VARIANCE COMPONENT ESTIMATION AND PREDICTION OF BREEDING VALUES

Abstract

A brief review is given of prediction of breeding values when the genetic and phenotypic variances of the trait are known. In many situations, however, these variances are not known and must be estimated from the data. A review is given of the traditionally used analysis of variance (ANOVA) type methods of variance component estimation and of the more recently developed methods of maximum likelihood (ML), restricted maximum likelihood (REML), minimum norm quadratic unbiased estimation (MINQUE) and some variations on MINQUE. Criteria for choosing among methods are discussed and the consequences of applying estimates obtained by these methods in place of known variances to the prediction of breeding values are considered. The methods are illustrated for estimation of paternal half-sib heritability and prediction of breeding values of sires using a small numerical example. Lastly, the consequences of selection on prediction of breeding values and estimation of components of variance for populations under selection are discussed.