Hello SPMers,
I'd like to add a question to a recent posting of Kristin McNealy to the
issue of extracting and adjusting time series for DCM.
We have studied the DCM chapter in the SPM5 manual and played through the
SPM2 example of Christian Gaser.
In this example, the model for DCM and the original GLM are identical,
that (at least it seems to me...) somewhat obscures the problem that in
other examples this may not be the case.
The scenario we face at the moment is having the 'original' GLM with
regressors A-D (of interest) and several nuisance variables. From this
model we get derive the peak voxels etc.
However, the DCM we set up has different input regressors: one is A-D
taken together, the second is B and D collapsed into one, and the third is
A and C collapsed into one. All the nuisance variables we left away so
far.
So the question - possibly dilletantic from a statistical point of
view...- is: if we want to adjust for the 'effects of interest', is it
equivalent if we cover regressors A-D (that is, usde the the original SPM)
or if we cover the mentioned combinations in the DCM model?
Really thanks for any recommendations here,
Philipp
|