Dear Todd,
I have (another)
question regarding residual motion effects in fMRI. I am wondering
if it is possible to put the 24 movement regressors into the K matrix of
the spm_filter routine (along with the high pass filter cosine functions)
to remove the residual motion effects rather than include them as
regressors in the design matrix? In my study, I have to keep the
number of regressors limited (<20 so I can apply a process that will
test for serial correlations in the residuals). I am comfortable
tinkering around with spm_filter, but I just want to make sure that if I
remove the residual motion effects this way, I am doing it
correctly. Can you provide any suggestions regarding
this?
Yes this is perfectly fine - in fact you could put all your confounds
into K;
and the estimation and inference will not change.
The only think you have to ensure is that the construction of the
residual-forming matrix based on K properly accounts for the fact
that
the confounds are not unit ortho-normal (i.e., X0' = pinv(X0) for the DCT
drift terms but this is not generally true).
With very best wishes,
Karl