| Resumo: |
Abstract In the final steps of a corn (Zea mays L.) breeding program, the test of the superior crosses is a standard practice involving yield trials in many test environments. The objectives of this research were to compare both the magnitude of the standard error of prediction of three different linear mixed models where location and hybrid effects were considered random or fixed effects and the magnitude of the phenotypic variance. Mixed, GLM, and Varcomp procedures from SAS were applied to grain yield and moisture data obtained from 40 hybrids evaluated at 22 locations within the U.S. Corn Belt. Results showed that when using a linear model which considers both location and hybrid effects as random effects, rather than fixed effects, standard errors of prediction were 33% smaller and differences could be detected with more accuracy. The three SAS procedures computed the same estimates of variance components using balanced data, but with unbalanced data, only Mixed and Varcomp procedures computed the same estimates. Phenotypic variance of grain yield was 52% smaller in Iowa environments, 9% smaller in Indiana environments, and 54% bigger in Illinois locations, compared to Nebraska environments where the hybrids were made.
|
