Estimating type B genetic correlations with unbalanced data and heterogeneous variances for half-sib experiments

• Published in: Forest science Vol. 45; no. 4; pp. 562 - 572
• Main Authors: LU, P.-X, WHITE, T. L, HUBER, D. A
• Format: Journal Article
• Language: English
• Published: Bethesda, MD Society of American Foresters 01-11-1999
• Subjects: Adaptation to environment and cultivation conditions; Agronomy. Soil science and plant productions; Biological and medical sciences; Fundamental and applied biological sciences. Psychology; Genetics and breeding of economic plants; Varietal selection. Specialized plant breeding, plant breeding aims; Agronomic character; Statistical method; Theoretical model; Genotype environment interaction; Genetic improvement; Mathematical model; Population genetics
• ISSN: 0015-749X; 1938-3738
• DOI: 10.1093/forestscience/45.4.562

A statistical approach is proposed for estimating type B genetic correlations with unbalanced data and heterogeneous variances across environments. First, parental GCA (one-half of the parental breeding value) effects are predicted separately for each environment using best linear unbiased prediction (BLUP). Second, predicted parental GCAs are weighted by their prediction accuracies and Pearson correlation between weighted GCAs in two environments is obtained. Third, type B genetic correlation is calculated by dividing the Pearson correlation with the mean products of prediction accuracies. Advantages of this approach include: (1) fixed effects are removed using best linear unbiased estimation (BLUE) so that they less seriously confound genetic effects when data are unbalanced; (2) heterogeneous variances associated with genetic group means are adjusted for during the process of parental GCA prediction; and (3) it is applicable when experimental designs are different in two environments. Numeric comparisons of estimation methods using simulated data of known true genetic parameters indicated that the new approach produces less bias and higher precision when data are highly unbalanced or have high level heterogeneity of variances across environments. For slightly or moderately unbalanced data, the method of Yamada (1962), however, is simpler and yields satisfactory estimates using standardized data. For. Sci. 45(4):562-572.