I'm having trouble determining whether the following is an acceptable,
albeit non-optimal, method for determining "psychophysiological
interactions" in a non-factorial experimental design.
We have PET images collected over two conditions: Rest(no drug) + Drug.
The "main effect" of the drug was first determined after proportionally
scaling the images (to account to the large decrease in global signal due
to the drug)
A region of interest was selected from this main effect analysis to serve
as the source region in the PPI analysis.
The adjusted activity within this region was extracted and passed into a
second level "single subject conditions and covariates" design. The
covariate was mean corrected across conditions (drug, no drug) and modeled
separately for each condition (3rd and 4th column of the design matrix).
The PPI was assessed by looking for the interaction using a contrast like:
[0 0 1 -1].
I'm concerned about the potential bias due to the "main effects". However,
I did apply an explicit mask of the main effects which seemed to not affect
the results of the PPI. Is this a valid procedure?
Thanks for your help in advance.
Nate White
|