OPTIMUM BIAS IN SELECTION INDEX PARAMETERS ESTIMATED WITH UNCERTAINTY

In the derivation of selection index weights it is typically assumed that population and economic parameters are known with certainty. In practice, however, estimates of selection index parameters must be used instead of the true parameters. It is shown that when errors in parameter estimates have asymmetrical effects on the efficiency of a selection index, the expected response from selection can be increased by biasing parameter estimates. In this way, the probability of making errors which result in large reductions in efficiency is reduced. A method of deriving optimum (biased) selection index weights when there is uncertainty in parameters is described. The method incorporates the error probability distributions of parameters estimated with uncertainty. In some examples, moderate (2-5%) increases in the expected response from selection occurred with uncertain heritability and economic weight estimates. Overall however, increases in selection response vary depending on the true index and are usually small (0 to 5 %) unless parameter estimates are extremely uncertain. Failure to account for uncertainty in unbiased parameters leads to over estimation of the value of selection. Applications of the method for both practical and theoretical purposes are discussed with specific reference to animal improvement programs.