Effects of correction for heterogeneity of variance on bias and accuracy of breeding value estimation for Dutch dairy cattle.

Data on 305-d milk yield from the Dutch dairy evaluation were used to obtain breeding value estimates from an animal model for 1984 to 1992. Changes in sire evaluations were investigated for bias and realized accuracy. Evaluations based on progeny records were generally lower than their expectation based on parent average. The average decrease was 157 kg for Black and White bulls and 73 kg for Red and White bulls. Evaluations based on test daughters changed about -50 kg when second and third lactations became available, but estimates for breeding value changed insignificantly when records on daughters from the breeding period of the bull were used. The standard deviation of changes from evaluations for first to second batch was about 14% larger than expected from population parameters. Breeding values of imported bulls, based on Dutch data, decreased 75 kg when more information became available in subsequent evaluations. Heterogeneity of variance was estimated by a quasi-likelihood approach with a model that accounted for sampling variance on estimates of variances within herd. The coefficient of variation of the variances within herd-year was 31%. A simple method for standardization of variances within herd-year decreased bias of parent averages by about 20%, and fluctuations of breeding values were within the expected range. A correction for heterogeneity of variance within herd may not remove all bias of parent averages, but a general improvement of bias and accuracy of breeding values can be expected.