Generalized canonical correlations and their applications to experimental data

Abstract

INTEREST HAS RECENTLY BEEN DEVELOPING IN EXPERIMENTAL DESIGNS WHERE WE HAVE SEVERAL OR MORE SETS OF MEASURES ON THE SAME SET OF ENTITIES, AND WHERE THE PROBLEM IS TO DETERMINE TO WHAT EXTENT THE VARIOUS SETS ARE ESSENTIALLY THE SAME OR DIFFERENT. The entities may be persons, tests, or whatever we wish. Suppose we have two or more sets of tests which we assume measure the same set of abilities. In particular, these may be sets of comparable forms for some specified battery of tests. We administer these sets of comparable forms to a sample of persons. The problem now is to determine to what extent the various sets are all measuring the same set of functions. A similar situation arises when we administer the same set of tests two or more times to the same set of persons. The problem here is to determine to what extent the same set of functions are being evaluated by the repeated administrations. (Author)