Interpreting canonical structure and scores in discriminant analysis
Oct 20, 2015 3:25 PM(1402 views)
I'm trying to interpret the relative contributions of my variables in a discriminant function analysis of chemical data for three groups of rocks. In a stepwise, linear, common covariance discriminant function analysis I'm able to generate non-overlapping 95% confidence ellipses for my three groups with no incorrectly assigned data points using 6 variables. My understanding is that the scoring coefficients are indicative of the partial/unique contribution of each covariate to the canonical function. The total canonical structure (aka the structure coefficients) indicates the correlation between each covariate and the canonical function. Is that correct and if so, how would it apply to the canonical details below?