Dear Soyoung,
Using the psychological main effect to identity the seed region will not
produce invalid
results but may be very inefficient.
As you rightly indicate; this selection renders the main effect of the
physiological
response correlated with the main psychological effect (and implicitly
their
interaction). This means that the regressors, which must include
both main
effects and the (PPI) interaction will explain each other; and any test
for one will
be much less significant than in a more orthogonal design
matrix.
The usual procedure is to identify the seed region using an orthogonal
psychological
contrast so that the physiological activity in the seed region can
interact with the (A>B)
psychological effect to explain the activity in regions showing an
interaction. Clearly,
this requires a factorial design with at least two factors (one to form
the psychological
effect and one to identify the seed region), which you may or may not
have.
I hope this helps,
With very best wishes,
Karl
At 14:09 23/03/2011, you wrote:
Dear Prof. Friston,
I have a short question concerning the orthogonality of a PPI
analysis
and would appreciate it a lot if you could give me some advice.
In our study, we have two conditions A and B.
The seed region for the PPI was defined by comparing A > B.
We use A > B as the psychological factor in the PPI analysis.
We include the PPI regressor (A>B*BOLD in seed), the
psychological
regressor (A>B, convolved) and the physiological regressor (BOLD
in
seed) in the PPI model.
My question is whether it violates the orthogonality assumption to
use
the same contrast (A>B) to define the seed region AND to use it as
the
psychological factor because the psychological regressor is now
obviously correlated with the physiological regressor (because the
former was used to find the latter).
Again, we would very much appreciate your feedback.
With best wishes
--
Soyoung Park
Affective Neuroscience
Freie Universität Berlin
Habelschwerdter Allee 45, 14195 Berlin
Raum JK 24/221e
Tel.: +49 (0)30 / 838 - 57168
eMail:
[log in to unmask]