A Monte Carlo Analysis of Some Alternative Estimators of Family Genotypic Values

Abstract

The selection problem in most animal breeding situations is the identification of replacement animals from a population composed of half-sibs, full-sibs or (as in pigs) litters. Often, any one population is composed of a mixture of these relationships pd, depending on the mating system practised and the population (or herd) size, a wide spectrum of relationships can be found. Information from such situations can best be in one of three ways, namely: selection entirely on family basis, saving only the animals from the best families (based on some measureh,. selection based entirely on individual performance saving only the best individuals regardless-of the performance of others in the family; or selection on some basis which combines emphasis on both the individual performance and the overall family performance. Broadly, the first alternative falls under family selecticd, the second under mass (or truncation) selection while the third is the subject of index selection. In mass selection the observation on the individual (candidate) forms the basis for selection. Family and index selection, however, involve computation of single values from individual observations. Strictly, family selection, in whatever form, is simply a special case of index selection in which either each family (or genetic group) gets its own index value which is then used to rank them or ranking is based on an index value computed for each individual within a family using the individual's own observation and some function of the overall family performance. The objective of this study was to investigate, using a Monte Carlo simulation, properties of three measures (or estimators) which can be used to rank genetic groups (or families), viz. simple average of genetic groups; average of genetic groups transformed to unbias the variance, and selection index derived from a one-way (genetic group) model.