Abstract:
|
Canonical correlation is one of the least used of the multivariate methods. It is the thesis of this paper that it is seldom used because it is incomplete and gives little information by itself. However, it can be highly illuminating when used to create a linked context of two or more multivariate spaces that in and of themselves have interesting internal structure. That internal structure can be experimental (with a MANOVA analysis within each space), a time series pattern linked across the spaces, or even just loose internal groupings by a collection of exploratory categorical variables. Also, the linking across the multidimensional spaces can be causal (e.g., mutual fund performance as predicted by market indices), or merely parallel (e.g., convergent behavioral and physiological measures of performance on cognitive tasks). Demonstrations are given of multivariate graphs for each type of internal structure, both in causally linked spaces and also parallel spaces.
|