a technique of multivariate analysis (specifically, two-stage principal components analysis) by which two groups of observed variables are transformed so that (a) the transformed members within each group (canonical variates) are linear combinations of the original variables, (b) the simple correlation between the canonical variates is maximized, and (c) successive pairs of canonical variates are orthogonal with preceding pairs —see orthogonality, nonorthogonal data This definition last updated 07/17/2008