Comparison of estimation and prediction methods of genotypic means in maize variety trials

The objective of this study was to assess different statistical approaches to estimating or predicting genotypic means of maize varieties in lattice experiments. The following models were evaluated: fixed model (FF), mixed model with random block effect (AF), mixed model with random treatment effect (FA), random model (AA), and shrinkage James-Stein estimator (JS). Forty-seven experiments were analyzed, each one with three replications and 15 to 28 treatments. The mean of two check cultivars (controls) per growing season was used as reference for the selection. In most experiments, the rate of genotypes selected by the shrinkage approaches (FA, AA and JS) was lower than by the FF and AF models, which also tended to select low-yielding genotypes, even when the genotypic determination coefficient (h2’) was low. At high h2’ levels the genotypes selected by the different approaches were quite coincident, although the ranking differed.