|
Abstract:
|
In conducting multivariate analysis of variance (MANOVA), Rencher & Scott, 1990, recommend using the four multivariate tests of significance to screen the univariate tests, as an effective method for controlling alpha inflation. One would expect that the vector configuration of dependent variables in such an analysis would have no effect upon the power of the multivariate tests, but such is not the case. Using Monte Carlo simulation methods a variety of between/within multivariate data configurations are created, half of them with Toeplitz vector patterns for the dependent variables, and half with clustered vector patterns. For three dimensional data configurations (and to a lesser extent for two dimensional) the multivariate significance tests are found to be somewhat less powerful at the low end (S/N ratios less than 1/1) when vector patterns are Toeplitz than when they are clustered.
|
