Hi
I think I understand better now.
When you analyse as seperate sessions, you get a result for session A that
you lose when you combine both sessions together. As you say, the most
likely explanation is increased error variance in session B that weakens
the effect.
Sorry not to be more helpful
Cathy
PS see PS below.
>Hi Cathy: thanks for your reply. I think I didn't express myself clearly
>enough. I am not talking about the actual conjunction analysis, but about
>the separate analysis of each task effect that precedes it. Yes, I have
>four different conditions (a different control for each task condition). I
>first analyzed the two paradigms (sessions) separately, with robust
>results. But when I analyze them as described below, the results are
>different and much weaker. In other words:
>
>Session 1 (= Paradigm A) +1 {using a user-defined hemodynamic model}
>on its own yields different results from
>
>Session 1 (= Paradigm A) +1
>Session 2 (= Paradigm B) 0
>
>I have a hunch that this may have something to do with greater variance.
>When I set the contrast weight to 0 for session 2 (in order to examine task
>effects for paradigm 1), would voxel intensities from session 2 nonetheless
>increase the overall variance? Btw, the problem is not solved by covarying
>for global intensity.
>The reason why I haven't even proceeded to do the actual conjunction is
>that I wouldn't expect adequate results from the conjunction of two
>individual contrasts that don't come out well on their own, so to speak.
>Thanks!
>Axel
>
>
>>Dear Axel
>>Re:
>>>I have a question regarding conjunction analyses in SPM96. The question is
>>>quite simple, but I haven't been able to find an answer in the archives.
>>>I have two task-control comparisons and would like to identify voxels
>>>activating conjointly during both. To be able to do a conjunction analysis,
>>>I understand, I first identify task effects for each paradigm individually.
>>>Since I use a custom 'hemodynamic' model, there's only one weight per
>>>contrast:
>>>
>>>Session 1 (= Paradigm A) +1
>>>Session 2 (= Paradigm B) 0
>>>to identify task effects for Paradigm A
>>>
>>>Session 1 0
>>>Session 2 +1
>>>to identify task effects for Paradigm B.
>>>
>>>This, I believe, corresponds to Cathy Price's advice to Andre Guillemin
>>>(12-18-99) in a similar context. What troubles me, though, is that the
>>>results are dissimilar from those I get when I analyze each session
>>>separately. I.e., my Z scores are substantially lower and the Z peaks occur
>>>in different loci. All data are intensity normalized to a common mean.
>>>I would be very grateful for suggestions on what might be the problem.
>>>Thanks!
>>>Axel Mueller
>>>
>>
>>It isn't possible for you to have higher Z scores for each of the
>>individual contrasts than you do in the conjunction if you are taking the
>>values from the same voxel. This is because in SPM96, the conjunction is
>>the sum of the two contrasts (with voxels excluded where there is a
>>difference between them). It would be more likely that the peaks in the
>>individual contrasts differ from one another and are therefore not included
>>in the conjunction. If the Z scores you are referring to in the
>>conjunction and individual contrasts all come from the same voxel, let me
>>know and I will think more deeply about it.
>>
>>Just to check, I presume you had four different conditions. Two for
>>paradigm A and Two for paradigm B?
>>
>>Cathy Price
PS I made a mistake here. The conjunction is the average of the the two
contrasts, not the sum of the two contrasts. Must have been rushing on
friday night.
Cathy
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|