The Invalidity Of Partitioned-U Tests In Canonical Correlation And Multivariate Analysis Of Variance

Publisher Website
Google Scholar
Abstract
The nearly universal practice of testing the statistical significance of individual discriminant functions and coefficients of canonical correlation by treating the residuals after successive terms of Wilks' lambda = U = -M Σ[SUP9,SUBi=1]. In (1-r[SUBi,SUP2]) have been subtracted as if the individual components were independently chi-square distributed is shown to be logically and statistically invalid. Two suggestions for valid, but conservative, alternative procedures are considered.

This publication has 0 references indexed in Scilit: