Hi Zheng
I did not receive any response yet, so I write what I figured out (and
if anybody proves me I am wrong I would be happy to hear that):
* there have been some discussions in the SPM-group how to use
2nd-order-models in order to calculate contrasts. My basic statistical
knowledge is not very high developed, but I understood that it is
problematic to put two 1st-order-contrasts into one 2nd-level-model, and
that the use of a constant in this model might lead to wrong results. In
my case, the contrasts are also assumed to be dependent.
* since I would not normally put my 1st-level-contrasts into one model
but test them in separate 2nd-level-one-sample-t-tests, I assume that
this approach is valid. Therefore, I took the spmT_xxx-images from my
two contrasts and used ImCalc to calculate a conjunction against
conjunction-null. This is done easily by using min (1, 2) > xxx, where
xxx is the cutoff t-value. The problem is that this approach does not
allow all the nice features SPM provides in the results section,
including visualization options. Fortunately, my conjunction did not
yield interesting results.
* I still do not understand why SPM allows only to calculate
conjunctions on contrasts that are included in the same 2nd-order-model.
For testing cognitive conjunction hypotheses, this seems to me a rather
unpractical approach.
I hope this helps.
Best,
Dominik
zheng wrote:
> Dear Dominik:
> I have almost the same question as yours. I didn`t find
> any reply in the list.Have you received any valueble answers?If so,could you
> do me a favor to send some to me?
> Meanwhile,what are the p values you used in conjuction
> analysis and the 2 component analysis?
> Sorry for bring your more questions but not answers.
> Thank you very much.
>
> zheng
> Shimane university,Japan.
>
> ----- Original Message -----
> From: "Dominik R. Bach" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Thursday, November 23, 2006 1:39 AM
> Subject: [SPM] Conjunction analysis with 2nd-level-t-test
>
>
>
>> Dear all,
>>
>> I have a problem when doing conjunction analysis with 2nd-level-t-test.
>>
>> We ran a design with 3 conditions: A, B, and C (control). In the first
>> level, we calculated (A - C) and (B - C) contrasts and analysed the
>> resulting con0001 and con 0002 images in separated 1-sample t-tests on
>> the second level. In order to perform a conjunction analysis, testing
>> against conjunction null, I understand that I must bring these two
>> contrasts into the same model. I therefore used a 2-sample t-test where
>> I put all con0001 in group 1 and con0002 in group 2. For conjunction
>> analysis, I define two contrasts, [1 0] and [0 1], and analyse them
>> together.
>>
>> However, when I analyse these contrasts alone, the results are quite
>> different from the one-sample t-tests we used before. I wonder why this
>> is so, and I assume that a conjunction analysis won't be valid in this
>> case. Must I therefore do this manually using ImCalc?
>>
>> Thank you for any input
>> Dominik
>> ---
>> Dominik R. Bach
>> University Hospital for Psychiatry
>> Bern, Switzerland
>>
>
>
>
>
>
|