Following up on my last e-mail, I noticed that the whitening matrices are
the ones causing different TSS's for different models.
After a dive on the list's archives, I found that this is probably due to
different covariance estimates, which in turn originate different W's.
I'm now facing the challenge of choosing the "most correct" W in order
to "level" the TSS values regardless of the model. For a random voxel, I
plotted the whitened time-courses for each model (see attachment).
Notably, the whitened time-courses are very similar for two of the models,
while in the remaing model there seems to be a negative offset of the
1) Wouldn't one expect the covariance estimates to be more reliable?
2) How does this mismatch in W propagate to error structure and what are
the consequences of this?
3) Is it OK to use the two out of three criterion to select a more correct
Thanks again and sorry if my questions are just plain naive.