Hi all
If I may add to the discussion, I noticed that in some models estimated with 3dDeconvolve the estimate of the constant term of the baseline model has a range of values which is quite different from the original epi data, even taking on highly negative values, possibly due to some degree of collinearity in the model (I posted a message on the board about this one or two weeks ago). If this indeed happens, normalizing with the constant term of the baseline model may yield quite unpredictable results. I would be really curious about whether anybody else has noticed this, and the possible explanations/workarounds: even though I began by normalizing by the estimated constant term, since it seemed theoretically more correct, I ended up using the grand mean across the brain of the entire epi scan as a normalization factor.
ciao
giuseppe