Dear Nicolas,
> > I tend to leave the stats questions to the stats people, who
> > know a great deal more than myself on the subject.
>
> Well, this is a dilemma, because the stats people do not respond :-(
> But perhaps my question was just too trivial or too VBM-specific.
>
> > > 1.) In SPM5, does it still make sense to use GM (or WM, respectively)
> > > integrals as a covariate in VBM (which was often done in SPM2-VBM)
> >
> > See http://en.wikibooks.org/wiki/SPM-VBM#Globals for my views on such
> > matters.
>
> You state there that "spm_global is not a good way of computing "globals"
> for VBM, as it was intended to compute a kind of average brain intensity
> (by thresholding out "non-brain" voxels, and computing the mean of what is
> left)." Does that mean, the option of 'Global calculation --> mean' should
> NOT be used for VBM?
Yes. It is more rigorous to provide your own "globals" that have some proper
interpretation. Then you can say e.g. that you localised differences that
could not simply be explained by total grey matter volume differences.
>
> A question (for which I did not find an answer in your wikipedia article)
> is whether you still recommend to include GM (or WM) integrals for VBM as
> it was often done in SPM2-VBM-analyses? I got the impression, though, that
> SPM5 automatically uses globals, too, if I try to include these integrals
> as a covariate. Therefore I thought that it does not make sense to use BOTH
> globals and integrals. However, a solution may be not to enter the
> integrals as a covariate in an Ancova but instead to choose: Global
> calculation --> user --> global values, and then enter the integrals. Does
> that make any sense?
I'm not sure what options you get with the stats, but I think it is possible
to do a proportional scaling "global" correction using your own values. As
for the ANCOVA correction, these can simply be treated like any other
confounding covariate.
I don't actually know what is best. It is really up to the users to decide
what question they want to ask of the data. The global correction is
essentially a part of this question.
>
> Two last and minor questions:
> 1.) Is it o.k. to use an absolute threshold of 0.05 for analyses of GM- and
> WM- volumes and density?
It should be OK.
>
> 2.) If I compare two groups (two sample t-test) and in regressions I should
> choose "variance unequal", right?
I'm not 100% sure here. In principle, the variances may be larger in one
group than another, but I don't know how well the correction is SPM deals
with this.
>
> Many thanks for your reply and help in advance.
> Should these questions be of general interest, feel free to cc your reply
> to the list (may save you some further questions on the same topic ;),
Will do.
Best regards,
-John
|