Hi Anderson,
I neglected to follow-up on the effect size information for the two tests. Here is a short recap.
Using the same patient and control groups, FA data was analyzed initially using TBSS then subsequently examined using the tbss_sym FA script. The design and contrast files were identical, and the number of permutations was identical (5000).
At alpha .10 (fslview .90) significant differences occurred in both tests, but at alpha .05 there was only a significant between group difference in the tbss_sym script results. A cursory comparison of Cohen’d results suggests somewhat similar effect sizes for the both tests:
At p <.10 , using peak MNI coordinates from the cluster report to locate peak coordinates in each test, the ratio of TBSS (mean=1.9) to tbss_sym FA (mean=2.4) effect sizes at these coordinates is about 1.26, and TBSS has about a 10% greater effect size. I then tried using fslmaths and fslmeants to extract effect size data as follows:
For tbss_sym FA
fslmaths raw_tstat1.nii.gz –div 2.12 full_d_sym
fslmeants –i all_FA_symmetrised_skeletonised –m full_d_sym –o full_d_sym.txt
For TBSS
fslmaths raw_tstat1.nii.gz –div 2.12 full_d_TBss
fslmeants –i all_FA_skeletonised.nii.gz –m full_d_TBss –o full_d_TBss.txt
Is the usage of fslmaths and fslmeants above is correct? The value 2.12 is Cohen’s d for 18 subjects, 2 groups of 9. If the above usage is correct, then the mean FA using fslmeants for TBSS is 0.4842 and for tbss_sym 0.4881. In other word the effect sizes for both tests are very similar. If this is the case, why would tyss_sym have a moderate size cluster (48 voxels) survive at .05 while the TBSS analysis showed no significant difference between groups at .05?
All the best,
Charlie
AndersonAll the best,Regarding which test found more pronounced differences, you may want to use a measure of effect size for this, or maybe show graphically the values for each subject.Hi Charlie,This may be an unduly simplistic interpretation. You need to discuss these results with the principal investigator who is conducting the study, taking into account the structures where the differences appeared, and the role that these may have in whatever training the subjects did. And read what the literature says about the sort of experiment you are doing -- plasticity it seems (?).
On 5 August 2014 15:20, charlie <[log in to unmask]> wrote:
FSL pundits,
Using the same patient and control groups, FA data was analyzed initially using TBSS then subsequently examined using the tbss_sym FA script. The design and contrast files were identical, and the number of permutations was identical (5000). With an alpha level of .90 (0.10) between group differences (p <0.10, corrected) occurred in both analyses: in the TBSS results the expert or trained group showed greater left FA than controls in several structures; the tyss_sym analysis showed leftward FA of experts in some structures and hence relative rightward FA in the same structures in controls.
This suggests that left hemisphere dominance in experts is a contributing factor of outcomes for both tests, and FA results of the tests may be correlated. Is this a valid inference?
At .95 (0.05), only the tbss_sym FA test showed a surviving cluster (bilateral difference) that differed between the groups. Just considering the two FA tests, could it be inferred that the tbss_sym FA script is more sensitive? Or would it be more accurate to say that the bilateral difference (tbss_sym) is more pronounced than the absolute difference (TBSS)?
Any feedback would be much appreciated. Thank you