Denis,
> 1.
> Could I use the 'user-defined regressor' to emulate this option ?
> In this case, and because it's a multi sessions analysis, I suppose that
> the array of this no-interest variable as to be cut in respect of
> session'lenghts.
Yes, you can use this option. You will have to cut your regressor, as you
suggest, and enter the 'cut' parts for each respective session. SPM will
expect a vector that is the same size as each session, in terms of scan
number.
> Then I suppose that this covariate account for some variance of the
> data. But is this variance 'removed' from the contrasts [1 -1 ...] for
> which this covariate is set to 0 ?
The variance explained/modelled by this covariate is accounted for when you
evaluate new contrasts that don't explicitly weight it (i.e. when the weight
above the covariate is set to '0'. Even if you don't set it to zero, SPM
takes a zero as its default.
> And how could I know how munch this
> covariate account for ?
If you want to ask 'what is the effect of including this covariate in my
model', then you can do a simple F-contrast of the form 0 0 0 0 0 0 0 0...
1... 0 0 0 0 where the '1' indexes your covariate, and the zeros index all
other covariates. This would show you voxels where a significant account of
variance is modelled by the inclusion of this particular regressor - i.e.
where it makes a difference to the overall model fit.
If you're looking for quantifying this in terms of percentage variance
explained, I suppose you could hack together something like an R^2 value.
I'm not sure what measure is most appropriate, however.
> 2. Is it an other way do defined a 'covariates of no interest' ?
> (even if matlab programing is needed)
I'm sure there is. Programmers, over to you.
Best
Dave McG.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|