Dear SPMers,
We ran a covariate only analysis in SPM99 and I'd like to make sure I'm
conceptualizing the results correctly. Six scans were given on each of two
days before and after treatment with drug or placebo. Twenty-two of 23
patients who received the drug reported that the symptoms of their pain
disorder improved or stayed the same. Control subjects changed less, with
about equal numbers improving and getting worse. We wanted to see what
brain regions were more active when patients reported higher symptom
scores, and what brain regions were more active when patients reported
lower symptom scores. We did +1 and -1 contrasts of the symptom rating
covariate (subject-specific and centered). Since each patient had only one
rating per day, we felt we were modeling only symptom change across days,
and could not put session into the model without vitiating the covariate.
In a prior analysis of this same dataset, Amygdala and BA24 were reliably
deactivated by treatment (confounded with session order, unfortunately,
since the pre-treatment session always came before post-treatment
session). The problem is that while I hoped these areas would be
positively correlated with the covariate, which I viewed as meaning more
active when symptom ratings are higher, the result showed them strongly
*negatively* correlated with the covariate, and showed the positively
correlated areas to be areas the prior analysis found to be increased in
the post-treatment session. Since the post-treatment session generally has
lower values for the covariate, I'm confused. Can anyone set me straight?
Thanks Much,
|