Planned Unbalanced Designs for Estimation of Quantitative Genetic Parameters. I: Two-Way Matings

Abstract

A number of unbalanced and balanced two-way classification designs (modified North Carolina Experiment 2) were compared for estimation of quantitative genetic parameters. In the designs considered, maximum likelihood estimates are easily computed by an iterated least-squares method. Asymptotic variances, together with simulation results, were used to compare the designs with respect to estimation of the genetic parameters considered. Simulation results showed that asymptotic variances gave reliable values for the sampling variances of additive genetic variance (5A), heritability, and predicted gain from selection, but not for dominance variance (a2) or degree of dominance (d). Except when error variance is quite large, suitably chosen unbalanced designs can produce more efficient estimators of a, heritability, and predicted gain, on either an individual or family mean basis, than balanced designs. Balanced designs are preferred for estimation of a 2 and d. Among the designs investigated, unreplicated designs from which cD and residual error variances cannot be separately estimated are often best for estimation Of cA, heritability and predicted gain when the experimental unit and selection unit is an individual plant.