This is more of a comment than a question, but I thought it might be useful to others who are attempting the "Percent Signal Change" method as described in the "How To" on the subject.

I noticed than when I ran 3dDeconvolve on my normalized data, the resulting statistical map looked funky, with a halo of activation around the edges of the brain. My local expert told that this pattern was indicative of head movement during the task. There was also considerably less apparent activation in regions I expected the task to elicit, including visual cortex. When I reran the analysis without normalizing the data, the pattern of activation seemed more in line with my expectations, and there seemed to be considerably less movement artifact.

My guess is that by normalizing the data by the mean of the signal (and including outliers due to movement in this mean) I managed to wash out most of the interesting, task-related differences in the signal.

Does this make sense? Do you think it would be wise to exclude outliers in calculation of the mean prior to normalization, or normalize the data after rather than before conducting analysis?

Thanks--
JL