Dear Carolin,
I have applied the TFCE method but I soon stopped using it because it
did not provide consistently better results than standard permutation.
However, I may have been unlucky and have discarded it too soon. It
should also be mentioned that with standard permutation we look at
least at two tests (peak-level corrected and cluster-level corrected)
rather than one.
As in all permutation approaches, it is dependent on an assumption of
exchangeability of the observations. It is less sensitive to other
violations of assumptions so I would have no problem believing its
results.
I think the main drawback is that it is less well understood than a
standard permutation test, where peak-level and cluster-level have a
familiar interpretation. Also, it is unlikely that the parameters that
define the statistic have an optimal setting with uniformly better
performance irrespective of the characteristics of the effect (this
happens rarely in statistics). However, it may be more difficult to
tailor the statistic to an expected effect distribution. It is an
interesting approach.
Best,
Roberto Viviani
University of Innsbruck, Austria
Quoting "Reichert, Carolin" <[log in to unmask]>:
> Hi SPMers,
> I am using CAT12 to compare grey matter density between two groups
> (n=13, t-test for independent samples, all participants above 60
> years). I went through the analyses as described in the CAT12 manual
> and checked group differences for significance in the results table
> (no significant differences after FWE-correction at peak or cluster
> level). Afterwards I applied the TFCE method to my contrasts of
> interest (with the defaults of 5000 permutations, Freedman-Lane
> permutation method) and my results slightly changed - now, there
> were regions significantly differing between groups after
> FWE-correction. However, I couldn't find many papers using the TFCE
> method and thus my questions are:
> Are there specific cases for which the TFCE-method is appropriate or
> inappropriate?
> Do I have to consider specific requirements when using the TFCE-method?
> And are there any disadvantages using TFCE (beside time for calculation)?
>
> Thanks a lot for any advice!
> Best,
> Carolin
>
>
> ------------------------------------
> Carolin Reichert, PhD
> Centre for Chronobiology
> Universitäre Psychiatrische Kliniken Basel
> Wilhelm Klein Strasse 27
> CH-4012 Basel
>
> Phone: 0041 61 325 5057
> www.chronobiology.ch
>
>
>
|