Fixed and Random Contemporary Groups

Abstract

For genetic evaluation of dairy cattle, contemporary group sizes are often small, and information is lost when contemporary groups are treated as fixed. Treating contemporary groups as random recovers some information across contemporary groups but may cause bias in prediction of breeding values if a nonrandom association exists between sires and contemporary groups. Results from a recent study indicated that treating contemporary groups as random gave consistently higher correlations between true and predicted breeding values, in particular when a positive association existed between contemporary group and sire effects. The present study shows that those results cannot be generalized based on correlations between true and predicted breeding values and on biases in breeding values for random and nonrandom association between sires and contemporary groups. For nonrandom association, the 50% best sires were assumed to be associated with half of the contemporary groups. If the best sires were used in the best contemporary groups, accuracies were higher when contemporary groups were treated as random effects (called the random model) than when contemporary groups were treated as fixed effects (fixed model) because of a reduced bias and a larger number of effective progeny. However, if the best sires were only represented in the worst contemporary groups, the correlation between true and predicted breeding values for the random model could become negative. If a nonrandom association exists between sires and contemporary groups, the groups should be treated as fixed effects for practical genetic evaluations.