Dear Juanjo,
I don't think it matters that a1 and a2 are not near in time.
Session specific effects will have been modelled out by other
parameters in the GLM.
You should ensure, however, that you have counterbalanced the
order eg. a1 does'nt always occur in the first session - otherwise
you will have a time counfound.
Then all you need to do is get the a1-a2 contrasts for
each subject and do a one-sample t-test.
Best,
Will.
Juanjo Lull (UPV) wrote:
> Hi Will,
> Since the last time that I recieved a mail from you there's been a long
> while. I'm still thankful about that e-mail (the one that is bellow),
> that helped me a lot, clarifying my ideas.
> I have now got stuck with another problem that has been in my mind for
> very long but it didn't come to me so clear untill now!!
> In the e-mail that you sent last time I was asking about two conditions,
> audio1 and audio2. The subjects would hear in the first session to
> audio1 and in the second session to audio2. Therefore, the conditions
> were not near each other in near blocks. While the explanation you gave
> about subtracting the contrasts was very good, now there is another
> problem. I have re-read the contrasts part in the spm book and I would
> like some confirmation...You advised me to make a contrast that
> subtracted audio1-audio2 (i.e. if the first column was a1 and the second
> a2, then I would do a [1 -1]) so I could use that contrast as input for
> a second-level analysis. I think that, as audio1 and audio2 are in
> different sessions, performing audio1-audio2 is not possible, as the
> blocks from each condition are not near, and therefore the conditions
> are not directly 'subtractable'.
> My question here would be (if what I have written up to now is ok):
> could I do a subtraction in another way, i.e. performing a paired
> t-test? Would I be able to do an ANOVA with those data in case I
> couldn't do the 'direct' subtraction, as I guess? How should I perform
> this ANOVA if it's possible?
> Thanks in advance,
> Juan J. Lull
> PS: feel free about sending this to the mail list, if you consider it
> could be interesting for the rest of the people
> PS2:
> session1: R-A1-R-A1-R-A1-R-A1
> session2: R-A2-R-A2-R-A2-R-A2
> where R=rest, A1=audio1, A2=audio2
> Will Penny wrote:
>
>> Juanjo UPV wrote:
>>
>>> Hi SPM,
>>>
>>> I'd like to ask a doubt about my data and second-level analysis.
>>>
>>> I have two fMRI block-design runs. These runs consist of a
>>> basic auditory stimulation paradigm
>>> (Rest-Activation-Rest-Activation-...). 3 groups of people are stimulated
>>> with both runs. Everyone listens first to the audio1 and then to audio2.
>>>
>>>
>>
>>>
>>> Now I'd like to know how to extract 2nd-level info. For the first level
>>> I have extracted, for every patient/control two contrasts, each asking
>>> for activation for each session (the 'famous' con* images). Now here
>>> is, finally, my doubt: I want to know:
>>>
>>
>>
>> Firstly, for each subject in each group create the 'audio1'
>> contrast images. You can also create a 'diff' contrast image for each
>> subject
>> which would be 'audio1-audio2' (ie. difference in effect between
>> sessions).
>>
>>
>>
>>> - wether audio1 generates a greater activation in one group than audio2.
>>> This means wether audio1 stimulates different areas than audio2 and
>>
>>
>>
>>
>> For, say, groups 1 and 2:
>>
>> Design 1: Use a two-sample t-test design (this will be the 'second-level'
>> design). For group 1 enter the diff contrasts for all subjects in
>> that group. Similarly for group 2. Then do a [1 -1] F contrast to see
>> voxels where activations are different between groups. You can use
>> [1 -1] and [-1 1] t-contrasts if you're interested in the sign of the
>> difference.
>>
>>> wether audio1 makes the activation signal greater.
>>>
>>
>>
>> Here, for each group you'd do a one sample t-test on the 'audio 1'
>> contrast images (or, you could collapse across groups if you were'nt
>> interested
>> in group specific activities here).
>>
>>
>>> - the same, but measuring audio2 vs. audio1
>>>
>>
>>
>> here, you'd do a one sample t-test on the diff contrast images
>> and then use a t-contrast of [-1].
>>
>>
>>> -I'd also like to know if a group activates more than the other two, or
>>> at least one of the others, for audio1, audio2 and both.
>>>
>>
>>
>> perhaps initially the best thing to do would be to look at the
>> main effect of group on activation.
>>
>> To do this use a 'one-way ANOVA (without constant)'. For number
>> of conditions - select 3 - for your 3 groups. Then enter the
>> 'diff' contrast images for each group. Finally, a
>> [1 -1 0; 0 1 -1] F contrast will show any voxels where the differential
>> effect is varies over 'group'. Note that in this 'matlab' notation the
>> ';'
>> operator specifies a new line - this F contrast is on two lines.
>>
>> Then you can do t-contrasts like [1 -0.5 -0.5] to see if the effect in
>> group
>> 1 is larger than the average effect in the other two groups.
>>
>> But after doing the F-test - it's probably easier just to go to the
>> voxels
>> of interest (ie. those that are significantly active in the F-map and are
>> in an interesting area neurobiologically) - and plot the parameter
>> estimates to
>> see what the pattern is (eg. group 1 larger than the others).
>>
>> Hope this gives you some ideas.
>>
>> Best, Will.
>>
>>
>>> I think I now know how to perform a conjunction of the contrasts for
>>> audio1 and audio2 (thank you Pilar Salgado!!), but I'm now interested in
>>> the RFX point of view.
>>>
>>> Any help will be greatly appreciated!
>>>
>>> Juan J. Lull
>>>
>>> ______________________________________________
>>> Juan J. Lull ------- jualulno_at_upvnet.upv.es
>>> [MI - Medical Imaging Area]
>>> BET - Bioengineering, Electronics and Telemedicine Group
>>> UPV - Politechnical University of Valencia - Spain
>>> ______________________________________________
>>>
>>
>>
>> --
>> William D. Penny
>> Wellcome Department of Imaging Neuroscience
>> University College London
>> 12 Queen Square
>> London WC1N 3BG
>>
>> Tel: 020 7833 7478
>> FAX: 020 7813 1420
>> Email: [log in to unmask]
>> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|