This paper presents an application of AMMI models - Additive Main effects and Multiplicative Interaction model - for a thorough study about the effect of the interaction between genotype and environment in multi-environments experiments with balanced data. Two methods of crossed validation are presented and the improvement of these methods through the correction of eigenvalues, being these rearranged by the isotonic regression. A comparative study between these methods is made, with real data. The results show that the EASTMENT & KRZANOWSKI (1982) method selects a more parsimonious model and when this method is improved with the correction of the eigenvalues, the number of components are not modified. GABRIEL (2002) method selects a huge number of terms to hold back in the model, and when this method is improved by the correction of eigenvalue, the number of terms diminishes. Therefore, the improvement of these methods through the correction of eigenvalues brings a great benefit from the practical point of view for the analyst of data proceeding from multi-ambient, since the selection of numbers of multiplicative terms represents a profit of the number of blocks (or repetitions), when the model AMMI is used, instead of the complete model.
genotype × environment interaction; multi-environments experiments; multivariate analysis
