Conditions under which canonical correlation and redundancy maximization produce identical results

Abstract

Redundancy analysis was proposed as an alternative to canonical correlation (van den Wollen-berg, 1977). The following paper examines the conditions under which canonical correlation and redundancy analysis produce identical solutions. It is shown that identical results are possible when all nonzero eigenvalues of the correlation matrix of the criterion variables are equal. As a special case, when there is only one canonical root or when all criterion variables are uncorrelated, canonical correlation and redundancy analysis produce identical results.