Precisão e controle de qualidade em experimentos de avaliação de cultivares

Abstract

This study had as objective to propose a new approach for quality evaluation of variety trials for determination of cropping and use values (VCU), which considers three attributes simultaneously: magnitude of the residual variation, replication number, and genetic control of the trait under selection. It was also emphasized the need for using shrinkage estimators/predictors of genotypic values instead of unshrunk phenotypic means of varieties, i.e., the procedures should consider the genetic coefficient of determination of the traits, as well as the eventual heterogeneity of residual variance within varieties. Targeting an accuracy of 90%, it was concluded that Snedecor F test values associated to treatment effects in the analysis of variance should be above 5.0. The magnitude of genotypic variability of the traits is also involved in the F statistics. This means that the approach of fixing minimum values for replication number and maximum values for residual variation coefficient (CVe) is not sufficient. For traits related to yield (with low genetic coefficient of determination) the normally used replication number, between two and four, does not permit to reach the targeted accuracy, even if residual variation coefficients below 10% are aimed, and the experimentation is conducted on several sites and years. For that target accuracy it is recommended the use of at least six replications. It was also shown that shrinkage estimators provide more precise and reliable inferences concerning genotypic means of the varieties, and their use is encouraged.