Dear Desmond,
I think the reason is simply that permutation is not fashionable. In
biostatistics or statistics, other approaches are much hotter. You can
find regular comments on the rather limited applicability of
permutation techniques in comparative reviews of approaches.
In neuroimaging studies permutation approaches are applicable except
in single subject studies, and except in certain models based on
heteroscedasticity at the first level.
In my experience permutation always comes on top of everything else in
terms of power. There are several known reasons for this, the main one
being that it models the discrete nature of the data. In addition, it
is robust against violations of distributional assumptions. It is
trivial to ensure the same distribution across the volume by
preliminarly ranking the data voxelwise -- albeit at the price of some
power. If you have structural datasets, it is difficult to think that
anything else offers comparable inference.
When submitting manuscripts, I have --thankfully-- never had any
problems with having used a permutation technique.
Roberto Viviani
Dept. of Psychiatry III
University of Ulm
Quoting Desmond Oathes <[log in to unmask]>:
> Dear SPMers,
>
> I've been looking for alternative multiple comparison correction
> approaches and thought that SnPM sounded interesting. It looks like
> it hasn't taken off in a big way in published work and there is no
> update for SPM8. Any idea why that is the case? Am I missing some
> controversy on the method or is it perhaps just a general reluctance
> from GLM trained folks (like me) to adopt non-parametric stats?
>
> Thanks for any input.
>
> Desmond
>
> ----------------------------
> Desmond J. Oathes, Ph.D.
> Stanford Psychiatry Department
> email: [log in to unmask]
> url: http://www.stanford.edu/~oathes/
>
|