candisc

Functions for computing and visualizing generalized canonical discriminant analyses and canonical correlation analysis for a multivariate linear model. Traditional canonical discriminant analysis is restricted to a one-way 'MANOVA' design and is equivalent to canonical correlation analysis between a set of quantitative response variables and a set of dummy variables coded from the factor variable. The 'candisc' package generalizes this to higher-way 'MANOVA' designs for all factors in a multivariate linear model, computing canonical scores and vectors for each term. The graphic functions provide low-rank (1D, 2D, 3D) visualizations of terms in an 'mlm' via the 'plot.candisc' and 'heplot.candisc' methods. Related plots are now provided for canonical correlation analysis when all predictors are quantitative.