Hi Donald,
Thank you for your swift response. I definitely will dive into the
code for your toolbox - thank you for making that available on the
listserv. For my own sanity, I just want to confirm I did things
correctly. First off, the psychological terms were indeed defined as
1s and 0s, as in:
Condition A: 1 0 0 0
Condition B: 0 1 0 0
Condition C: 0 0 1 0
Condition D: 0 0 0 1
I constructed PPIs for each separately, and then combined them into a
single design matrix (incl. the seed regressor only once). Screenshot
attached ("new.png"). Once estimated, I used the following weights to
acquire the differential connectivity across A and B: [0 0 1 0 -1]
(padded with zeros for remaining columns, and replicated for the
second session)
Is this legit? And is essentially what your toolbox implements?
Just in case it's helpful, I've also attached a screenshot of a design
matrix using the "old" method ("old.png"). Here, the psychological
regressor is:
Condition A - Condition B: 1 -1 0 0
And here, I estimate differential connectivity of A and B by simply
weighting the 3rd column of each session (i.e. the coefficient for the
PPI regressor).
Many thanks for your help.
Bob
----------------------------------------------------
Bob Spunt
Doctoral Student
Department of Psychology
University of California, Los Angeles
On Thu, Mar 24, 2011 at 5:14 AM, MCLAREN, Donald
<[log in to unmask]> wrote:
> Bob,
> It's hard to say without being able to see the PPI regressors and how they
> were built in SPM5 (e.g. was the psychological term defined as 1s and and 0s
> or 1s and -1s?). The current implementation of gPPI uses 1s and 0s to build
> four separate regressors.
> I would use the gPPI code distributed a few weeks ago that has been tested
> (www.martinos.org/~mclaren/ftp/Utilities_DGM)
>
> In terms of results, I am not surprised by large differences on a complex
> task. The reason being is that in the traditional approach, you were lumping
> connectivity for C and D with the baseline connectivity (seed regressor) and
> assuming that both A and B are different from these. In the new approach A
> could be different and B,C,D, and baseline could all be the same and then
> you would find differences in A versus B.
> Let me know if you have further questions.
>
> Best Regards, Donald McLaren
> =================
> D.G. McLaren, Ph.D.
> Postdoctoral Research Fellow, GRECC, Bedford VA
> Research Fellow, Department of Neurology, Massachusetts General Hospital and
> Harvard Medical School
> Office: (773) 406-2464
> =====================
> This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
> HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
> intended only for the use of the individual or entity named above. If the
> reader of the e-mail is not the intended recipient or the employee or agent
> responsible for delivering it to the intended recipient, you are hereby
> notified that you are in possession of confidential and privileged
> information. Any unauthorized use, disclosure, copying or the taking of any
> action in reliance on the contents of this information is strictly
> prohibited and may be unlawful. If you have received this e-mail
> unintentionally, please immediately notify the sender via telephone at (773)
> 406-2464 or email.
>
>
> On Thu, Mar 24, 2011 at 3:28 AM, Bob Spunt <[log in to unmask]> wrote:
>>
>> I have a question regarding generalized PPI as has been advocated by
>> Dr. McClaren on this list. If I am understanding correctly,
>> generalized PPI simply involves estimating multiple PPIs in a single
>> model. In my case, I have four conditions: A B C D. Therefore, my gPPI
>> model should include 9 regressors (per session, and not including
>> additional covariates of no interest, such as motion regressors):
>>
>> 1 seed regressor
>> 4 psych regressors (A, B, C, D)
>> 4 PPI regressors (A, B, C, D)
>>
>> I would like to estimate the differential connectivity with a seed
>> region in the contrast A > B. I have already done this in the
>> traditional way with a psychological regressor based on the comparison
>> of interest (using SPM5), and just recently tried it using the method
>> described above. To be clear, here are the steps I took for each
>> subject:
>>
>> 1. Built PPI regressors (in SPM5) for each condition (against implicit
>> baseline) separately
>> 2. Estimate a some 1st-level model (for each subject) with the seed,
>> psychological regressors for each condition (4 total), and PPIs for
>> each condition (4 total)
>> 3. Generated a contrast image among the PPI coefficients for my task
>> comparison of interest (A > B)
>> 4. At the group-level, performed a one-sample t-test on participants'
>> A > B contrast images
>>
>> Using this method, I am observing dramatically different results than
>> with the traditional method. Is it plausible that these two approaches
>> would yield very different results, or is this likely an error in my
>> procedure?
>>
>> Please let me know if there is any other information I can provide,
>> and many thanks in advance for any advice.
>>
>> Cheers,
>> Bob
>> ----------------------------------------------------
>> Bob Spunt
>> Doctoral Student
>> Department of Psychology
>> University of California, Los Angeles
>
>
