The use of multidimensional scaling in analysing multi-attribute genotype response across environments

Abstract

Three-way Multidimensional Scaling is presented as a single analysis of genotype x environment x attribute data. The concept and underlying model of this technique are discussed, and its usefulness is investigated by applying the analysis to a well-known soybean data set. The measured attributes were considered to represent the line response in each environment. It was assumed the lines could be considered in an underlying space of r dimensions. The environments were able to perceive different response patterns in that different importances could be placed on these underlying dimensions. By applying this technique a spatial representation of the lines in a low-dimensional space was obtained. Thus a single multi-attribute analysis is achieved rather than attempting to combine separate analyses for each attribute. The resulting line response pattern was examined in relation to previous reports on this data set. The inclusion of all six attributes revealed information not previously obtained by separate analysis of two of these attributes. If general inference about genotype response is to be made rather than inference on one particular attribute, then the use of a technique such as this is recommended.