Hi -
Using TBSS/randomise I am looking at how FA and MD are associated with cognitive measures in a group of normal subjects (n > 150). My models all include one EV that is all ones (EV1) and then 1-6 other EVs of demeaned, normalized cognitive measures and/or demographic information. What is the correct way to do F-tests in these models?
Option 1 (3 EVs - the mean (EV1) and two covariates (EV2 and EV3). Each covariate gets its own F test):
contrast 1 [ 0 1 0 ] --> F1 [ 1 0 ]
contrast 2 [ 0 0 1 ] -- > F2 [ 0 1 ]
{Note: I know the F tests should be in columns not rows!}
Option 2 (3 EVs - as above, the mean and two covariates. A single F test)
contrast 1 [ 0 1 0 ] --> F [ 1 1 ]
contrast 2 [ 0 0 1 ]
I've been interpreting option 1 as representing F-stats for the individual covariates, i.e., where is EV2 (F1) or EV3 (F2) significant; and option 2 as representing both contrasts, i.e., where is either EV1 or EV2 significant. Is this correct?
I know this sounds like a basic question but I have been getting some very confusing results where massively significant corrected tstat images (e.g., the association of age with FA) produce an F stat image of all "0"'s when option 1 is used. Many of my covariates are correlated, but I do not orthogonalize the EVs to each other since they are all demeaned and normalized prior to putting them in the models.
Thanks for you help with this.
-Paul
|