Approximate accuracy of genetic evaluation under an animal model

Abstract

A procedure is described to approximate the accuracy of genetic evaluation by Best Linear Unbiased Prediction under an Individual Animal Model. This involved adjusting the diagonal elements of the coefficient matrix in the Mixed Model Equations for connections to parents, progeny and fixed effects. The reciprocals of the resulting coefficients then provided a good approximation to the diagonal elements of the inverse of the coefficient matrix, and, hence, approximate prediction error variances and accuracies could be calculated. Simulation was used to check the adequacy of the procedure. Type scores for 282 030 Canadian Holstein heifers were replaced by simulated records, maintaining the original data and relationship structure. Analysing this data and grouping animals according to the amount of information available, empirical accuracies, i.e. correlations between true and estimated genetic merit, were calculated within groups. Predicted, approximate accuracies were consistently higher than their empirical counterparts. Differences were small for animals in the data, amounting to 1 or 2% only. For parents not in the data, empirical accuracies were 92% or more of predicted, except for sires with 6–20 daughters. Diagonal elements of the inverse coefficient matrix were obtained iteratively for a subset of the data, and showed a correlation close to unity with the approximate values. Overall, the approximation procedure performed well and can be recommended for use in conjunction with the genetic evaluation of large populations under an Individual Animal Model.