Jason,
> > The Global Null method requires orthogonalization, while the
> > Conjunction Null method doesn't. Were you looking for
> > orthogonalizatoin for Conjunction Null inference?
>
> Thanks for the reply! No, I was just wondering which implementation of
> Global Null corresponds to the one described in the Friston et
> al "Conjunction Revisited" paper.
Yes. In fact, the Global Null of that Friston et al paper and the Nichols
et al Conjunction paper are one in the same. What the "Conjunction
Revisited" paper introduces is the intermediate null, where you try to
reject a null that k of the nulls are true, where 1<k<K and K is the
number of tests to be conjoined.
The SPM2 modifications in spm2_updates implement the Global, Intermediate,
and Conjunction Nulls.
> After reading other recent posts on conjunctions, my understanding
> is that it's "inappropriate" to apply the standard FDR correction
> (apparent in the SPM2 GUI) in the case of conjunctions. In this context,
> is multiFDR is the optimal approach or will FWE suffice??
No no! FDR corrected thresholds and P-values are valid for all of the
conjunction options. It is simply that I've found that the FDR results
for the Conjunction Null are not so sensitive. In previous posts I've
offered an alternative, intersection of FDR-thresholded images approach,
which makes sense but which has yet to be shown to be valid. But that
'under development' method isn't available directly from the SPM GUI.
Does this clarify things?
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|