Hi again,
> Thank you for your comments. I'm not sure I understand the statistics
> entirely but at least what I've found is not impossible. Is the
> 'variance component estimation' pooled over the whole brain? It seems
> very counterintuitive to me that changing a mask mainly over the
> cerebellum should have effects on stats in the parietal lobe - what is
> the rationale for this?
Hmm, no real "rationale" I would say. How about calling an "emerging property" from the way things are implemented. A bit like consciousness you know.
It is not pooled over the whole brain. Rather it is pooled over those voxels that seem to (p<0.001 uncorrected, I think. Or is it p<0.05 for Basic models. If not, maybe it should be? Will/Guillame?) have some relation to your design (as assessed by the effects-of-interest F-contrast). It really isn't an easy problem. If you don't pool the estimates for each and every voxel will be very uncertain (though correct in expectation). If you pool the estimates will have higher precision, but may be wrong in expectation.
I think I would personally lean towards thinking that your estimates for the second (larger) mask are maybe a little "better" since they are based on a larger population of voxels. But it is really anybodys guess as to the outcome of the precision/bias trade-off.
>
> I am attaching a small file illustrating the scale of the problem - my
> F values have gone down by 5% with the change of mask, which seems
> like quite a lot to me (the area I'm interested in is highlighted).
If you look at the F-contrast images (the ones actually depicting the contrast) you see that they are slightly different, which would support my initial guess for why this happens (i.e. variance components have changed).
I would personally very much welcome if the outcome of the var-comp estimation was easily interrogated through the SPM-GUI (maybe is in SPM5 ??) or that at least some low-dim information such as e.g. corresponding Greenhouse-Geiser factor was reported.
>
> More pragmatically, can anyone explain how the standard mask would be
> calculated if I had done my smoothing at the first level, rather than
> after estimation. Then maybe I can use that method to get a better
> mask. At the moment I'm just using masks generated by a different
> study (also spm2, whole brain, same acquisition parameters etc),
> because I was assuming it didn't matter.
Hmm, I am not sure that messing with the masks is really the way to go here. What you have is a "borderline" significant area, one that will or will not exceed your 0.05 threshold depending on the exact outcome of the pooled variance-component estimation. I would rather level with the editor/reviewer, report your p-values and say that depending on your exact choice of parameters it changes a bit. Just my 5c.
Good luck Jesper
|