Dear Marjolein,
> I chose the multi-subject: conditions and covariates design, for
> PET scan. The four scans/conditions (A,B,C,D) were entered in
> time-order. I want to compare only two conditions (e.g. B and C) in
> the analysis. Do I have to put the other conditions on 0 while
> defining contrasts (i.e. 0,-1,1,0), or do I have to make another
> spm.mat file in which only the two conditions I want to compare are
> taken and define contrasts as 1,-1? Is there a difference between the
> two ways?
The first solution is the more appropriate one. One should generally try
to specify one design matrix modelling all observations and then use the
one estimated parameter set to compute all the contrasts.
The difference between the two solutions is that in the first case you
make the assumption that it is valid to use all scans for estimating the
variance under the null hypothesis, even if a contrast vector element is
zero for the associated basis function/condition. This is a valid
assumption for a fixed effects PET analysis. As a result, you have more
degrees of freedom in your statistical test at each voxel such that the
analysis is more sensitive to the underlying signal.
Stefan
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|