Linda,
In the standard PPI approach, you are modelling the effect as being
+/-, meaning one has to go up and the other has to go down relative
after controlling for all other regressors. Now, if only one is
different from baseline OR you have more trials for one type and the
difference isn't symmetric; then the model will still try to be fit to
the data. However, it will be biased toward saying there is a
difference by matching the condition that has the most trials because
as that is how to minimize the model error.
When you model the conditions seperately, then you minimize the error
equally for both conditions irrespective of the trial numbers. Thus,
the model error is lower with gPPI. The one caveat is, as you pointed
out, that the accuracy of the estimate increases with increasing
trials.
Hope this helps.
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 Wed, Jul 6, 2011 at 10:33 AM, L. Geerligs <[log in to unmask]> wrote:
> In follow up of my previous post about the validity of my PPI results I have
> been looking
> in dept at the results of the generalized PPI and the traditional PPI
> analyses together with
> my colleague Remco Renken. We found a strong effect of condition (target vs.
> standards in an oddball tasks)
> in the traditional PPI analysis; from a seed in the precuneus we found
> reduced correlations in target compared
> to standard trails with the rest of the default mode network (including the
> precuneus) at p_FWE <0.05.
> However with the generalized PPI, we found no difference between conditions
> at p<0.001 uncorrected.
> When we looked at the different contrasts in the generalized PPI analysis
> (PPI of targets and PPI of standards separately)
> we found the same default mode network in both. However, the effect was much
> more significant in the standard than in the target
> condition due to the reduced variance in the standard condition compared to
> the target condition due to the large difference
> in the number of trials. Therefore, we now believe that in the traditional
> PPI analysis comparing conditions with unequal occurrences,
> the result is dominated by the effects in the most frequent condition.
>
> Further discussion on this topic would be much appreciated.
>
> Kind regards,
> Linda Geerligs
>
>
>
>
>
> MCLAREN, Donald wrote:
>>
>> I haven't done a lot of investigations with the seed region; however,
>> it is possible that the relationship within the seed region is
>> variable across conditions.
>>
>> As for the differences between the two methods (standard approach and
>> generalized approach). Did you include all three conditions in the
>> generalized approach? We have a paper under review that suggest the
>> standard approach shows more significant results; but that these
>> results might be to due to a bad model (e.g. you are trying to fit the
>> data to your model and the model is not correct). This conclusion was
>> driven by an analysis of the AIC being lower when the conditions were
>> modelled separately.
>>
>> What were your thresholds?
>>
>>
>> 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 Mon, Jul 4, 2011 at 9:37 AM, L. Geerligs <[log in to unmask]> wrote:
>>
>>>
>>> 1) After your question I have been looking at the VOI data again and I
>>> found
>>> that I did
>>> not correct for effects of interest properly when extracting the VOI.
>>> Therefore, I repeated the analysis today using the PPI toolbox of Donald
>>> McLaren
>>> using both the traditional PPI analysis as well as the generalized PPI
>>> analysis.
>>> In both cases I corrected for 'effects of interest' . The traditional
>>> analysis led to exactly
>>> the same result as before, showing decreased correlation in target trials
>>> compared
>>> to standard trials from the seed region to a large network including the
>>> seed region.
>>> The generalized analysis showed no difference at all between target and
>>> standard trials.
>>> I would greatly appreciate any suggestions on what is causing these
>>> effects.
>>>
>>> 2) The contrasts for the three conditions (standards, targets, novels)
>>> were
>>> -1 1 0
>>> (fixation was the baseline and was not explicitly modeled in the design).
>>> Other factors in the design were the movement parameters.
>>>
>>> Kind regards,
>>> Linda
>>>
>>>
>>>
>>> Darren Gitelman wrote:
>>>
>>>>
>>>> Linda
>>>>
>>>> 1) Did you adjust the data? (When you click the Eigenvariate button it
>>>> asks if you want to adjust the data.)
>>>>
>>>> 2) You say below that your PPI was [1 -1] for the 2 stimulus types,
>>>> but you list 4 conditions below (standards, targets, novels and
>>>> fixation). What did you include in your PPI and what were the other
>>>> columns of your design?
>>>>
>>>> Darren
>>>>
>>>> On Fri, Jul 1, 2011 at 12:01 PM, L.Geerligs <[log in to unmask]> wrote:
>>>>
>>>>
>>>>>
>>>>> Thank you for your fast reply.
>>>>> I actually first did the PPI using the standard method in SPM. Then I
>>>>> realized that the PPI and
>>>>> Y terms were correlated and I thought this might cause the results
>>>>> (some
>>>>> of
>>>>> the variance in
>>>>> Y being explained by PPI). Then I tried again with the orthogonalized
>>>>> version of
>>>>> the PPI variable but the results remained the same.
>>>>> The task contains three conditions (standards, targets, novels) and a
>>>>> fixation.
>>>>>
>>>>> Kind regards,
>>>>> Linda
>>>>>
>>>>>
>>>>>
>>>>> On 01-07-11, "MCLAREN, Donald" <[log in to unmask]> wrote:
>>>>>
>>>>> What happens when you use gPPI or PPI in SPM?
>>>>>
>>>>> The way it seems that you did your PPI is non-standard, so its hard to
>>>>> tell if its the method or something about the underlying activity that
>>>>> is making the results look the way they look.
>>>>>
>>>>> Do you have condition A, condition B, and fixation in your task?
>>>>>
>>>>> 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 Fri, Jul 1, 2011 at 11:06 AM, L. Geerligs <[log in to unmask]> wrote:
>>>>>
>>>>>
>>>>>>
>>>>>> Hi SPM users,
>>>>>>
>>>>>> I have a problem with the interpretation of the findings in a
>>>>>> psychophysiological interaction (PPI) analysis.
>>>>>>
>>>>>> Recently I did a PPI analysis on event related fMRI data in which one
>>>>>> event
>>>>>> was presented more frequent than the other (oddball task).
>>>>>> I selected a seed region by using the first eigenvariate of the time
>>>>>> courses
>>>>>> of all voxels in a 6 mm radius around a peak voxel.
>>>>>> Then I looked at the difference between the two stimulus types
>>>>>> (contrast
>>>>>> 1
>>>>>> -1).
>>>>>> The results of this analysis showed decreased connectivity from the
>>>>>> seed
>>>>>> region to a large scale network, in the less frequent condition
>>>>>> compared to the more frequent condition. The network we identified
>>>>>> made
>>>>>> sense according to existing literature.
>>>>>> The curious thing about the result is that we also observed a decrease
>>>>>> in
>>>>>> connectivity within the brain area which I chose as the seed.
>>>>>> Moreover, when I repeated this analysis with different seed regions, I
>>>>>> found
>>>>>> different networks, but similar decreases of connectivity
>>>>>> with the brain area from which the seed time course was constructed.
>>>>>>
>>>>>> Therefore, I started to wonder about the validity of the findings. Is
>>>>>> it
>>>>>> possible that a PPI analysis with conditions with
>>>>>> unequally frequent stimuli gives spurious results? And if not, how can
>>>>>> it
>>>>>> be
>>>>>> that I find a decreased regression from the seed region to
>>>>>> itself in one condition compared to another?
>>>>>>
>>>>>> The PPI model was constructed in SPM8, in such a way that the ppi
>>>>>> variable
>>>>>> was orthogonal to the Y and P variables (using spm_orth)
>>>>>> and the data was filtered with the first eigenvariate from the signals
>>>>>> of
>>>>>> white matter and csf voxels.
>>>>>>
>>>>>> Thanks a lot!
>>>>>>
>>>>>> Kind regards,
>>>>>> Linda
>>>>>>
>>>>>>
>>>>>>
>>>>
>>>>
>>>>
>>>
>>>
>
>
|