CHOICE OF AN APPROPRIATE BENDING FACTOR USING PRIOR KNOWLEDGE OF THE PARAMETERS

A new method for the choice of an appropriate bending factor in the construction of genetic selection indices was established and evaluated by Monte Carlo simulation. The main feature of the proposed method is to use that bending factor which maximizes the correlation between true and estimated aggregate genotype, replacing the (unknown) population parameters through guessed values ("priors") in the computation formula. The efficiency of this and some other procedures relevant to index selection was investigated on altogether 336 different parameter and sample situations with 1000 replicates each. It was found that even in the case, where the assumed reliability of the used priors was pretty poor, the efficiency of the proposed method compared to all other procedures investigated was astonishingly high. For example, using priors equivalent to estimates originating from a half-sib analysis with a sample size of 400, the suggested method was on average only 3.3 per cent less effective than the best possible bending procedure from theory. On the other hand, compared to an unadjusted index, the observed improvement was 20.1, 8.7 and 2.2 per cent, when the sample sizes were 400, 800 and 1600, respectively. A further essential advantage of the proposed method is its self-adaptation to changing parameter and sample conditions. This property allows its universal application to any form of index selection. In all, the possible improvement of index selection by bending turned out to be quite substantial, whereby, in general, the improvement is the larger the more unfavourable the conditions for the construction of genetic selection indicates are.