Statistical Analysis of Yield Trials by AMMI and GGE: Further Considerations

TLDR

Recent reviews have compared the merits of AMMI and GGE models for yield‑trial data, highlighting the importance of selecting the most predictive model within each family to accurately delineate mega‑environments. This review addresses more than 20 issues requiring clarification after controversial statements and contrasting conclusions in recent reviews. The AMMI2 display captures more genotype main effect and GE interaction than GGE2, yielding a more accurate which‑won‑where pattern for complex datasets; when a single principal component captures GE interaction, AMMI1’s nominal‑yield display more clearly identifies winning genotypes and adaptive responses than GGE2, and within a single mega‑environment, a simple mean‑vs‑stability scatterplot is preferable to the GGE2 biplot.

Abstract

Recent review articles in this journal have compared the relative merits of two prominent statistical models for analyzing yield‐trial data: Additive main effects and multiplicative interaction (AMMI) and genotype main effects and genotype × environment interaction (GGE). This review addresses more than 20 issues that require clarification after controversial statements and contrasting conclusions have appeared in those recent reviews. The AMMI2 mega‐environment display incorporates more of the genotype main effect and captures more of the genotype × environment (GE) interaction than does GGE2, thereby displaying the which‐won‐where pattern more accurately for complex datasets. When the GE interaction is captured well by one principal component, the AMMI1 display of genotype nominal yields describes winning genotypes and adaptive responses more simply and clearly than the GGE2 biplot. For genotype evaluation within a single mega‐environment, a simple scatterplot of mean and stability is more straightforward than the mean vs. stability view of a GGE2 biplot. Diagnosing the most predictively accurate member of a model family is vital for either AMMI or GGE, both for gaining accuracy and delineating mega‐environments.