Hi Rick:
So yeah actually the WM/CSF was not put through an initial regression, nor was the actual data as there was no 3dDeconvolve step at all - it was just never run in this paper I'm replicating. I wasn't actually planning on running 3dDeconvolve at all because of this very reason (i.e. it not being done in the initial manuscript) but after your reply and it seeming as if 3dTproject can use output from 3dDeconvolve if you run it with the -x1D_stop, it seems like I probably should just run a 3dDeconvolve with the standard motion parameters.
As to the WM/CSF, it seems as if I was originally mistaken. Instead of my WM/CSF nuisance signal being a 3D+time file, it is instead a matrix of principle components (i.e. a voxelXcomponent matrix - 500,000x5, for example) so running the file in 3dTproject using the -dsort option will no longer work. As I still need my output file to be a 3D+time dataset, can I still run 3dTproject but make my WM/CSF matrix into a 1D file and use it with the -ort option?
Follow-up question. In the 'notes' section it says, "* If A is a matrix whose column comprise the vectors to be projected
out" - which I believe is true here - "define the projection matrix Q(A) by
Q(A) = I - A psinv(A)
where psinv(A) is the pseudo-inverse of A [e.g., inv(A'A)A' -- but
the pseudo-inverse is actually calculated here via the SVD algorithm.]
Would I have to define Q(A) manually by calculating the pseudo-inverse of A in some outside program (and if Q(A) DOES need to be calculated by the user, what does 'I' represent)? I'm assuming not because of the very next line. Is this note just technical info about how A is projected out or does it indicate something I have to do myself?
I ask all this assuming 3dTproject is still the tool I should be using given the structure of my new nuisance file. Is this true?
Sorry to bother you. Your help has been great and is much appreciated.
Lauren