There are differences between SPM's method of handling ANOVAs and those of
conventional stats packages. (The ones I've heard about in the past are
ReML instead of e.g. Greenhouse-Geiser (for sphericity), and pooled
instead of partitioned errors (as far as I can tell SPM can give
partitioned errors, but only with extra work by the user, even in the case
of SPM5).)
Another thing stats packages do, however, is correct for multiple
comparisons, in this case the fact that most ANOVAs will likely have many
tests. While of course the number of tests is tiny compared to the usual
source of multiple comparisons in neuroimaging (the huge number of
voxels), SPM doesn't actually do this other more conventional multiple
comparison correction, does it?
|