I'll try to answer the first two questions.
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping)
>[mailto:[log in to unmask]] On Behalf Of Mike Angstadt
>Sent: Monday, October 23, 2006 11:52 AM
>To: [log in to unmask]
>Subject: [SPM] a few PPI questions
>
>Hi,
>
>I have a couple of questions about performing a PPI analysis in SPM2.
>
>First, what are the reasons to adjust/not adjust the data when
>extracting a VOI? This post by Darren Gittelman suggests not
>adjusting the data
>http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind03&L=SPM&D=0&I=-
>3&m=12246&P=426684
>
>But the sample analysis from Will Penny and several other
>posts on here suggest adjusting for Effects of Interest.
My response may have been in error as far as what the processing stream
"should be", but because of quirks in how spm_regions interacts with
spm_peb_ppi, it turns out to be fortuitously a correct answer. My answer
today would be adjust for the effects of interest, but here is why choosing
<don't adjust> also works BUT only if the intention is to do PPI.
WARNING: boring description of matlab code follows
On line 91 of spm_regions the code sets up the menu that allows you to
choose what to adjust. Eventually a popup menu appears which includes
<don't adjust>
effects of interest
... other F contrats
A corresponding variable q is also setup, which will index the appropriate F
contrast in the xCon field (the structure array that contains all the
contrasts). The first entry of q is 0, the second is usually 1,
corresponding to the effects of interest.
If you choose <don't adjust> then on line 101, xY.Ic is set to 0.
Things roll merrily along until line 173. Since xY.Ic is 0, the program does
not enter that if statement, but resumes processing on line 187.
On this line it sets up xY.X0, which will contain the whitened confounds of
the block/session effects (SPM.xX.iB) and the mean column (SPM.xX.iG).
This gets exported into the VOI file.
Now you press the PPI button and import your VOI files. On approx line 204
of spm_peb_ppi the code removes the whitened confounds from Y (the first
eigenvariate). This effectively removes the mean and the session effects
from Y (i.e., very similar to having adjusted for the effects of interest).
For illustration if we take a single session standard design (i.e., no user
entered regressors) the two ways of processing the data return very similar
results.
1) if you DO adjust for the effects of interest the results are
raw data - whitened confounds (mean column) -> first-eigenvariate |then in
spm_peb_ppi first-eigenvariate - whitened confounds.-> Yc.
2) If you DON'T adjust for the effects of interest the results are
raw data -> first-eigenvariate |then in spm_peb_ppi first-eigenvariate
- whitened confounds.-> Yc.
Thus it all generally comes out very similar in the end.
-------------------------------------------
>
>Second, when setting up a PPI in SPM, what's the difference
>between not including a particular task from my model, or
>including it with a weight of 0?
None. Setting the weight to 0 effectively removes it from the model.
--------------
Paul Fletcher has already done a nice job of answering the 3rd question. So
I'll stop now.
Darren
>
>Third, and perhaps most important. When looking at a PPI on
>two tasks, A and B, how do we interpret positive and negative
>PPI results? My understanding from the Friston paper
>(Neuroimage, 1997) and from discussions I've had with others
>is that a positive PPI between two tasks is showing an
>increase in the correlation between the seed area and the area
>the result shows up, while a negative PPI result shows a
>decrease in the correlation.
>My confusion comes from two things. First, is this
>increase/decrease based on the absolute value of the
>correlation (so an increase is actually saying they're more
>significantly correlated and a decrease is less significant)
>or the raw correlation (so going from 0 to 0.9 would be an
>increase, but so would going from -0.9 to 0). And the second
>source of confusion comes from some published papers that
>report positive PPI results as positive correlations (and
>negative PPIs as negative correlations) rather than changes in
>correlation between tasks.
>
>Thanks
>
>-Mike
>
>Mike Angstadt / Senior Research Coordinator University of
>Chicago Medical Center / Department of Psychiatry
>5841 South Maryland Avenue, MC 3077
>Chicago, IL 60637
>Phone: 773-834-5942 / Fax: 773-834-4536
>Email: [log in to unmask]
>
|