Dear Hamzah,
in order to directly compare the parametric SPM and the non-parametric permutation test in the TFCE toolbox, you should rather use the non-parametric T-test (and not the TFCE). If these T-tests reveal comparable results the effects you have noticed are due to the nature of the TFCE statistics. It enhances clusters (also below the threshold) and acts like a filter that pushes large distributed effects as well as high focal effects.
Best,
Christian
PS: Your link only works with our login...
On Thu, 20 May 2021 16:50:19 +0100, Hamzah Baagil <[log in to unmask]> wrote:
>Hi,
>
>I am analyzing 3 groups using VBM CAT12 toolbox. The F-test showed differences at FWE < 0.05 for SPM parametric and TFCE test and the area and the cluster size is comparable for both tests. When I continue to do post-hoc test however the SPM and TFCE test showed a rather discrepancy regarding the results, when for example comparing Group A and B (please see link). What I dont understand is that for the TFCE test the difference between Group A and B is much bigger compared to Group A and C both corrected at FWE < 0.05, whereas for the SPM parametric test at FWE < 0.05 I saw cluster surviving for the comparison for Group A and C, but not corrected for Group A and B (only uncorrected < 0.001 at FWE cluster level < 0.05). The 2 images from the link shows the comparison between Group A and B for the TFCE test with many clusters surviving FWE < 0.05 and the second image the same comparison with the SPM parametric test at p < 0.001 cluster level FWE < 0.05.
>
>https://ukaachen-my.sharepoint.com/:f:/r/personal/hbaagil_ukaachen_de/Documents/CAT?csf=1&web=1&e=HBV4BG
>
>When I do two sample t test separately between the groups the results for SPM parametric tests and TFCE seems to be more comparable and also reasonable.
>
>Any help would be appreciated.
>
>Best,
>Hamzah
|