Statistical test of genotype-by-environment interaction patterns observed from a biplot

This short article is to comment on the methodology and conclusions of Yang et al. (2009) concerning the use of biplots to reveal crossover genotype-by-environment interaction (GE) patterns, which was published in Crop Science. Yan et al. (2007) applied GGE (genotypic main effect plus genotype-by-environment interaction) biplot analysis to an Ontario winter wheat dataset and concluded that there were two megaenvironments for winter wheat in Ontario, i.e., eastern Ontario represented by locations E5 and E7, and southwestern Ontario represented by seven other locations (Fig. 1), Yang et al. (2009) took this study as an example of “over-utilization or abuse of biplot analysis.” Yang et al. (2009) applied a bootstrap procedure to the same dataset and concluded that the crossover GE pattern revealed by the biplot analysis was false. They advocated that any future use of biplots should be supplemented with a statistical test, such as the bootstrap procedure they introduced. We believe their conclusion and recommendation are unfounded and misleading.