Strategies in Canonical Correlation with Application to Behavioral Data

Publisher Website
Google Scholar
Abstract
Since canonical correlation is being increasingly applied in the behavioral sciences, a comprehensive appraisal of its merits is warranted. A survey of the literature employing this technique indicates incomplete use of the method and confusion in canonical terminology. This paper reviews and integrates four analytical procedures that provide interpretive assistance for canonical solutions: (a) testing for significance, (b) estimating stability, (c) naming dimensions, and (d) predicting within dimensions identified. These procedures are used to analyze 309 skilled factory worker job value and perceived job characteristic responses to 35 work referent items. Results show that: (a) only three of the twelve highly significant (p < .01) canonical correlations indicate meaningful underlying constructs, (b) stability estimates are necessary to identify these constructs, (c) only a selected subset of work referent items is crucial to dimensional naming, and (d) prediction is bi-directionally significant (p < .01) within each dimension interpreted. Future canonical applications can provide substantive contributions to the behavioral sciences only with a full appreciation and implementation of these analytical strategies.

This publication has 18 references indexed in Scilit: