hello all ~

i have a question about the mechanics of partialing variance in the LVs produced by SVD in PLS.  it holds that the following phenomenon can occur in such analyses: if groups/conditions A and B with multiple levels within each are analyzed together, the first LV picks up on a 'common' variable (e.g., common effect of RT in a behavioral analysis) - if then, these are analyzed separately the differences between levels within each group might 'move-up' to LV1 (i.e., they might have been present in LV2 or 3 of the combined analysis).i understand that this occurs because the components are produced in order of covariance.

what is the mechanism, however, that accounts for the situation where a given effect between levels is significant in the within-group analysis, but was NOT significant int he combined analysis. is this because the amount of covariance accounted for by this effect in the combined analysis will be smaller and more variable, thus, making the singular value not-reliable via permutation?

thanks for the clarification!
agatha