Dear Dr. Gitelman and others,
The design is factorial. We have task/control X 2/3/4 choice (multiple 
choice questions). 2x3 design.

When getting a VOI, it asks to adjust for effects. I chose "Main effect 
of task", which would be the F contrast (temporal derivs included):
 >> SPM.xCon(2).c'
ans =
     1     0     1     0     1     0    -1     0    -1     0    -1     
0     0
     0     1     0     1     0     1     0    -1     0    -1     0    
-1     0

[This would be 2task TD 3task TD 4task TD 2control TD .....]

I assumed this would be analagous to section 33.3 #5 in the SPM5 manual 
(pp. 287). Then I chose a sphere of radius 4mm.
For the particular region I was asking about, it is in IFG/insula, which 
I know is probably pretty heterogeneous; but I also see the same thing 
when doing a PPI with BA6 (right in the center of FEF) as the seed 
region, and also in the superior parietal lobe.

RE the variance: assuming the formula in line 283 of spm_regions is 
correct, for a random subject:
 >> xY.s(1)*100/sum(xY.s)
ans =
   95.8826
(lowest of all subj's was 88.5%; the rest were 91.3% or higher)

So, would Hyoung-Ryul's ideas make sense? Would this be something to 
explore in DCM (modulation of self-connections for those regions)?
Thanks,
Chris

Darren Gitelman wrote:
> Dear Chris
>
> I agree that is is not usual to see an activation at the site of the 
> source region. PPI knows nothing about self or other connections per 
> se so this doesn't explain the finding (although we could hypothesize 
> various causes for the variance in that region) .  I wonder if your 
> eigenvariate is not a good representation of the activity in that 
> region so that the interaction term (PPI.ppi) is picking up nearby 
> voxels.  When you extracted the VOI, how many voxels were included in 
> it and how much variance did it represent? Also you say you adjusted 
> for the main effects of task. Do you mean you included all task main 
> effects (using an effects of interest type of contrast that only 
> excluded nuisance effects) or that you adjusted the eigenvariate to 
> only include variance from 1 particular task. The latter is generally 
> not recommended (i.e., the eigenvariate should not be restricted to a 
> single task as the source of variance but should include all real 
> experimental effects).
>
> Darren Gitelman, MD
>
>
> 2010/3/17 강형률 <[log in to unmask] <mailto:[log in to unmask]>>
>
>     Just to add an idea: maybe there is a task-related self-connection?
>     In other words, the seed region's activity might have shown more
>     quadratic autocorrelation during the task than during the control.
>
>     PPI.Y can be different from the raw time courses of the BOLD
>     signal of the voxels in the seed region because it is the first
>     eigenvariate of them.
>     Thus, the portion of the variance that correlated with PPI.ppi
>     would have come from the rest of the variance, after excluding the
>     first eigenvariate.
>
>     Best regards,
>     Hyoung-Ryul.
>
>
>     On Thu, Mar 18, 2010 at 1:25 AM, Chris Watson
>     <[log in to unmask]
>     <mailto:[log in to unmask]>> wrote:
>
>         Hello,
>         In doing a group PPI, I see that in the results there is a
>         significant cluster at the seed region I chose (in this
>         example, BA9). Does this result make sense? [This is from an
>         fMRI experiment; task vs. control]
>
>         From this page
>         (http://www.fmrib.ox.ac.uk/Members/joreilly/what-is-ppi) in
>         Figure 1, it is stated:
>         "Voxels in which activity is equally correlated with the seed
>         region timecourse all the time will not show any correlation
>         with the PPI regressor."
>
>         Also in an email from Dr. Penny
>         (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&D=0&P=1605413
>         <https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&D=0&P=1605413>):
>         "if the beta value for I is significantly non-zero there is a
>         psycho-physiological interaction ie. the psych variable
>         changes the correlation between source and sink voxels."
>         The psych variable shouldn't change the correlation between
>         the seed region and itself.....
>
>         I followed the steps in the manual (1. Get VOI in seed region,
>         adjusted for main effect of task; 2. Do the PPI; 3. Specify +
>         estimate design, w/ regressors PPI.ppi, PPI.Y, PPI.P). I would
>         have thought that the PPI.Y regressor would explain the
>         variance in the seed region...
>
>         Any thoughts?
>
>         Thanks,
>         Chris
>
>
>     -- 
>     Hyoung-Ryul Kang, M.D.
>     Seoul National University, Functional Brain Imaging Laboratory
>
>