Dear experts,
a question about which regressors to include for a PPI analysis.
If I understand correctly, and following Will Penny's example on the SPM
website,
(http://mail.google.com/mail/?ui=1&view=page&name=js&ver=7u9e09ods4ep)
when carrying out a PPI one should include 3 regressors: 1) the
main effect of your psychological variable; 2) the main effect of your
region, i.e. the timecourse at the seed ROI; and 3) the PPI, or the
interaction between the two.
My question is whether it is sufficient to include just these 3 regressors
or whether one should include the *other* task-related variables
explicitly modelled in your experimental design. It occurs to me that in
the absence of these regressors the 'psychological' variable may capture
not only variance associated with that task, but with all other tasks with
which it is non-orthogonal. this is probably not a problem if you have a
factorial design, and your psychological variable is one of the main
effects (e.g. [1 1 -1 -1] for a 2 x 2 design) but what about the case
where the contrast specified comprises only a subset of all trials (e.g.
[1 -1 0 0])? My guess is that under such circumstances one should include
task-related variance corresponding to unmodelled factors as well when
constructing the design matrix for the PPI in order to be able to
explain any results specificially in terms of functional connectivity
owing to that psychological factor...?
Or perhaps a similar appraoch as often adopted in DCM studies, i.e.
including instead a 'photic' regressor that captured variance associated
with a potentially confounding nonspecific psychological process (such as
simple visual stimulation) would be another option?
your thoughts are much appreciated
Chris
Christopher Summerfield
[log in to unmask]
|