A comparison between the linear regression and AMMI approaches to study multi-site trials using a sample of South African maize trials

Abstract

Grain yield for 21 maize cultivars common to 10 localities and three harvesting seasons was used to compare the Finlay-Wilkinson and the additive main effects and multiplicative interaction (AMMI) methods of studying interactions in multi-site trials. Both methods showed variability and unpredictability with respect to year-to-year results. Each method considered the data from different points of view, bringing to light various aspects of the variability contained in the data. The cultivars were loosely ordinated by the AMMI method along a gradient representing length of growing season in two years out of three and also when the years were analysed jointly. In order to maximize the predictive ability of the AMMI method, selection should be made using the AMMI 0 scale of treatment estimators because of its minimal sampling variance.