Iris
Simply think that you make a design matrix and each regressor (column)
has its own variance (nothing shared) - so if you have 1 column for a
variable of interest it may show some activation profile A; now if you
add another variable and this one has for some voxels a correlation with
the 1st variable (it will always occurs unless you explicitly
orthogonalize your variables) then for the 1st variable you'll end up
with a profile B, probably similar to profile A if the regressors aren't
too correlated - this profile B is like A minus the shared variance.
back to your question now ..
> Dear all,
> I want to include both a covariate of interest and a nuisance covariate in my
> design matrix (2nd level analysis; one sample t-test; SPM5). My question
> regards the contrast specification.
> The design matrix has 3 columns: 1st column: con images of subjects; 2nd
> column: covariate of interest; 3rd column: nuisance covariate.
> What is correct (t-contrasts): 0 1 0 or 1 1 0 (to see activations)?
> Is it possible to look at both contrasts, since 1 1 0 gives me far more
> significant results than 0 1 0? (I fear that 0 1 0 is "correct".)
> How will I calculate deactivations (0 -1 0 or -1 1 0)?
>
T: [1 0 0] is the contrast you want to see areas activated taking into
account the effect of the other covariates (no matter of interest or not)
T: [0 1 0] will show areas that correlate with you regressor of interest
having removed the effect of group (and the variable of no interest)
F: [1 0 0; 0 1 0] is the combination of the above ; I don't think that T
[1 1 0] is really meaningful ...
Best
Cyril
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|