Hi,
> I have 4 contrasts arising from a second level analysis. First, I wanted to
> know what is the common regional activation for all of the four contrasts.
Yes - that requires a 'conjunction null' analysis - as you probably
saw from the Friston-Nichols papers.
> For that I defined an ANOVA including all for contrasts. Then I choosed all
> four contrasts to calculate a conjunction analysis (using the "conjunction"
> button). I got nice results at a p<.001 uncorrected level. O.K.
Bear in mind of course that you couldn't use this to infer that the
results were 'significant' in any meaningful brain imaging sense,
given that you have used an uncorrected threshold.
> But
> afterwards, I tried an "interaction analysis". For the contrasts A, B, C
> and D I built a contrast as described by "1 -1 -1 -1" to get the exclusive
> part of activation represented only(!) by contrast A.
So, this is just A - B - C - D. I suspect you wanted to get something
more like A - (B + C + D)/3 - which would be something like 3 -1 -1 -1
> I did the same for
> the other contrasts. What is SPM concretely doing in this case? Does it
> pool B, C and D? And when doing so, how does it do this? Is there a good
> reference to this?
I'm afraid you would have to read the basic stuff on the GLM in SPM in
order to understand this. I wrote an old page which might help:
http://www.mrc-cbu.cam.ac.uk/Imaging/Common/spmstats.shtml
> It seemīs to be a trivial question but Iīm convinced that many community
> members are curious to get more knowledge about this. To the time it seemīs
> not possible to publish fMRI data without using these new fashion
> procedures.
On the contrary, I would say it will cause you less trouble to do the
straightforward contrasts...
Best,
Matthew
|