Multivariate-Covariance and Canonical Analysis: A Method for Selecting the Most Effective Discriminators in a Multivariate Situation

Although univariate analysis of grouped data can provide valuable information it often cannot discriminate between groups. In the analysis of grouped data a method is required which combines the advantages of multivariate analysis with economy in relation to the number of variables required. To attain this objective two computer programs were used. The first was concerned with the selection of a subset of the original variables so that no significant between-group variance remained. The second did a canonical analysis. These programs have been used on laboratory data for a gilgaied soil sampled randomly within three microtopographic classes, based on contour, at four depth intervals. Multivariate analysis has confirmed the hypothesis that significant differences in the pattern of soil properties occur between the microtopographic positions. Moreover, the economization procedure has also suggested the variables most closely related to group structure.