Diallel designs, analyses, and reference populations.

Some issues which evidently remain as areas of contention in diallel analysis are reviewed. In the estimation of genetic variance components for an ancestral reference population in equilibrium, analysis should be applied to F1 data only, and proceed as for a random effects model. No meaningful reference population is constituted by the parental sample itself, but an equilibrium reference population which could be derived from these parents can be defined. To give unbiassed estimates of parameters for this population, analysis must include S1 data, and the parents must be homozygous and in linkage equilibrium. Estimation of dominance variance must allow for the fact that the diallel population itself is not in linkage equilibrium by correcting the specific combining ability sum of squares for the mean S1 vs. F1 difference. Some different analyses of variance of the diallel table are discussed in the context of the above reference populations.