Jon,
> I have been following the thread about SnPM and low df neuroimaging
> studies with some interest. I have downloaded the SnPM toolkit
> and am beginning to experiment with it - I am still at a bit of
> a loss about the optimal number of permutations to estimate a model.
As many as possible, within reason :)
If the total number of possible permutations is less than 5000, then
I use all of them. If there are more than 5000, I use an approximate
test with just 5000 permutations.
Note that even an "approximate" test still is exact (it appropriately
controls the Type I error rate), but it might not be as powerful as using
all permutations.
> Is this covered in your new HBM paper?
Yes. The copyright agreement with Wiley allows me to email a PDF to you,
but I can't globally post it.
SPMlist: If anyone else would like a copy, just email me.
-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
|