Hi
I am comparing (H2O) PET images between a placebo and a drug condition.
In this context, I also want to analyse the correlations of activity
(perfusion) with psychometric variables. For this I use a covariate
without interaction (the data of the covariate appear in one line
beneath each other in the design matrix).
The problem is that some of those covariates are specific for the drug
condition, with (almost) zero values in the placebo condition.
Now, when computing a design containing both the placebo and drug
images and the covariate for both conditions (the placebo ones being
almost zero), the result is almost identical with the result obtained
from computing a design consisting of the placebo condition alone. When
computing the analogous design consisting of the drug condition alone
with its covariate the result looks very different.
Why does the placebo condition apparently dominate the drug condition
in this setup? Is it - as I would assume - due to the near-zero values
of the covariate in the placebo condition?
Is there any approach to include the differences of the images
(perfusion) between placebo and drug conditions without distorting the
analysis by the near-zero covariate in one condition?
Or is it best to just compute the correlations of the perfusion with
the covariate in the drug condition alone?
Thanks a lot and best regards,
David
--
David Andel
Neuropsychopharmacology and Brain Imaging
Psychiatric University Hospital Zurich
Spamtrap: [log in to unmask]
|