Hi Henk,
The relationship between t and F applies to single voxels. When it comes
to TFCE, which a statistic that combines the height of the original t or
F with the spatial extent, that original relationship is lost.
Moreover, the t-test has only one tail, whereas the F-test is
two-tailed. This changes not only the peaks, but also the support area
that defines the spatial extent of TFCE.
> Am I right in concluding that the use of non-parametric tests reduces the sensitivity of the analyses when using F-tests?
No no, at least not with respect to this observation using TFCE. What
you observed is the (expected) effect of TFCE, in which the relationship
between F and t is no longer the same as with the original statistics
that don't have spatial support, and this all has nothing to do with
permutation tests.
All the best,
Anderson
|