Okay thank you so much for the help. One thing I noticed (which originally turned me off from the PE's) occurred when I calculated a "voodoo correlation" just to see if the method was working. So for example I had a task and than looked at the association between voxel activity for that task and treatment outcome (medication). Then I wanted to see how activity in those voxels during the task were associated with response time (a third variable). I extracted mean pe's for the voxels whose task processing activity was associated with treatment outcome. To check that this method was working I then calculated a partial voodoo correlation in which i correlated mean pe per subject in that mask with treatment outcome (the criteria that had been used to select those voxels) I expected a very high correlation but found that it was very low. I than extracted z-scores instead of PEs and got a high correlation. I assumed that this was because the z-scores account for within subject variability. Is this not the case?
|