Handling outliers in multi-environment trial data analysis: in the direction of robust SREG model

ABSTRACT

Site regression model (SREG) is utilized by plant breeders for the analysis of multi-environment trials (MET) to examine the relationships among test environments and genotypes (G) and genotype-by-environment interaction (GE). In its regular form, singular-value decomposition (SVD) is applied on residual matrix from one-way analysis of variance (ANOVA) to partition G plus GE effects. However, ANOVA and SVD are sensitive to atypical observations, which are common in MET. To overcome this problem, three robust models are proposed to obtain valid results even in the presence of outliers. Their efficacy was evaluated by simulation and compared with standard SREG. Different scenarios were considered to identify the appropriate strategies to deal with outliers in real situations. Two real datasets are also presented to highlight the usefulness of the proposed methods in agricultural data. Our results indicate that the use of the proposed alternatives enables to effectively analyze MET data in the presence of outliers and maintain good performance without them as well.

KEYWORDS:

• Multiplicative models
• multivariate methods
• outliers
• robust approach
• site regression

Acknowledgments

The results presented in this work have been obtained by using the facilities of the CCT-Rosario Computational Center, member of the High-Performance Computing National System (SNCAD, MincyT-Argentina).

Disclosure statement

No potential conflict of interest was reported by the author(s).