Hello all,
I have a few basic questions regarding design matrix and contrasts to be
used in randomise on TBSS data. I am very grateful for any input and
suggestions.
I want to correlate various electrophysiological and behavioral measures
with FA of each subject of every voxel in the skeletonised mask. I also
want to regress out the effect of sex and age. That is, I am not
interested in sex differences per se.
Q1: Would this then be a sensible design matrix?
[Group: 1=female, 2=male]
[EV1:female][EV2:male][EV3:age][EV4:behavior][EV5:electrophys]
Q2: Should the continuous variables (EV3, EV4 and EV5) be demeaned or in
any other way be normalized prior to running randomise? If so, would this
be taken care of by applying the -D flag running randomise?
Q3: If I'm interested in the effects of one EV on FA, with the effects of
the other EV's regressed out, would I have to apply different contrasts in
order to test for both positive and negative correlations? I.e. something
like:
0 0 0 0 1 (to yield voxels with a significant positive correlation with
EV5) and
0 0 0 0 -1 (to yield voxels with a significant negative correlation with
EV5),
or am I getting something wrong here?
Q4: I'am also interested in interactions between different EV's. Can this
be addressed directly at this level in randomise? If so, how would
contrast(s) testing for significant interactions between i.e. EV3 (age),
EV4 (behavior) and FA in each voxel look like?
Thank you very much.
All the best,
Lars
|