Dear Souheil,
> I've modified spm_fMRI_design.m to not detrend the columns of my design
> matrix. So now I've got a design matrix where the columns do no have
> zero mean and are slightly non-orthogonal. I've also taken out the
> main session effect.
>
> I can go and estimate the model and generate beta images and everything
> seems okay.
>
> If I then want to go and do statistical comparisons, is there something
> I need to watch out for?
>
> ie if I have block design experiment and I put in a regressor for task
> and a regressor for the baseline (each of which is a box car with a
> mean > 0), then the comparison task-baseline (contrast [1 -1]) should
> be the same as if I had left spm_fMRI_design.m unchanged and only put
> in a regressor for task with the contrast [1 0].
>
> Is the above correct?
Yes it is. The only thing you have to be aware of is that some
contrasts may include the constant term (or session effect) which is
clearly greater than zero for fMRI. Usually any contrast is precluded
from testing for a constant by including the constant explicitly in the
design matrix. If it is removed certain linear combinations of
regressors may be collinear with the constant and be very signifcant.
There is no problem with this, as long as these contrasts are not
mis-interpreted.
I hope this helps - Karl
|