Nonparametric statistical methods for analysis of genotype × environment interactions in plant pathology

Abstract

Host genotype × environment (GE) interaction is a commonly observed phenomenon in experiments in plant pathology and can be analysed using different statistical models which can include parametric or nonparametric methods. Using data based on ordinal scales (commonly used to measure disease in plant pathology) or continuous scales, the disease responses of host genotypes across environments and resulting GE interactions have been analysed using 26 different nonparametric methods. An analysis of two datasets for severity of gray leaf spot of maize (10 northern or southern adapted genotypes planted at 10 or 11 locations in the US, respectively) revealed differences among host genotypes, environments and the GE interactions. A factor analysis was performed on the rank correlation matrix arising from the application of each nonparametric method. The 26 methods can be categorized, for both sets of experimental data, in two major groups: (i) those which are mostly associated with level of disease severity and show little or no correlation with the static type of stability concept, and (ii) those in which both level of disease severity and stability of performance are considered simultaneously to reduce the effect of the GE interaction. This analysis separated nonparametric methods based on a dynamic concept of stability from those which are based on the static type of stability. Comparison of original and corrected nonparametric statistics indicated that, the identification of the most resistant cultivar for a specific environment based on both genotype and GE interaction would be useful to plant pathologists since estimates based only on genotype and environment effects are insufficient. Following measures based on corrected Ketata’s rank, original Ketata’s rank, Fox’s statistic and rank-sum, the NP2 statistic of Thennarasu (average of summation of absolute values of adjusted ranks from the median) was found to be useful in detecting the phenotypic stability of disease severity in a dataset. Nonparametric methods can be useful alternatives to parametric methods and allow valid conclusions to be drawn while assessing the GE interaction in experiments undertaken in plant pathology or plant breeding.