Linear regression and stability measurements as possible aids in wheat selection trials

Abstract

The effectiveness of the linear regression approach to the handling of genotype X environment interactions was evaluated with respect to irrigated and dryland spring wheats, intermediate and winter wheats and spring barley. Joint regression analysis of yield data suggests that the linear model does not satisfactorily explain the differential responses of genotypes to different environments, implying a more complex cause for such interactions than is provided for by the model. It was not possible to find a better fit of the model with regard to sub-groups of genotypes exhibiting similar responses, nor was it possible to propose biological causes for the sub-groupings in all cases. For the majority of genotypes it was not possible to improve on the linear regression by fitting other curve types. Stability and adaptability parameters estimated from linear regressions are therefore expected to have little predictive value under local conditions. Stability measures based on partitioning of interaction sums of squares have limited value in yield selection programmes because of the necessity to obtain extensive observations.