Hi FSL Experts,
I have what seems like a very basic a problem with several possible causes, though each cause is likely to be very simple.
I'm using randomise to run a simple f-test with white matter FA (via TBSS) across three unequal subject groups. The t-stat maps from randomise -- both corrected and uncorrected -- have minimum and maximum values of (±)0.000000, and most, but not all, have peak voxels of 255 0 0. The f-stats, however, have nonzero (if small: F=9.04, corrp=.237) maximum values at interpretable locations.
Even if there were no significant groupwise differences in FA, it seems implausible that max t-values would be 0.000000, much less if f-stats are nonzero, even if not significant. Is this normal or even axiomatic behavior for randomise when performing an F-test in randomise? And what are the likeliest causes & fixes for this behavior if it's not just a matter of interpretive failure on my part?
I have run various other t-contrasts on this dataset with error-free, interpretable results. I only seem to encounter these issues with f-contrasts.
Input/Output below. NB: groups are low, mild, and severe scores on a subscale of the Childhood Trauma Questionnaire (CTQ).
CTQ.mat:
/NumWaves 3
/NumPoints 75
/Matrix
1 0 0
1 0 0
...
0 0 1
CTQ.con:
/NumWaves 3
/NumPoints 3
/Matrix
1 0 0
0 1 0
0 0 1
CTQ.fts:
/NumWaves 3
/NumPoints 1
/Matrix
1 1 1
randomise options: -i all_FA_skeletonised -o CTQ -m mean_FA_skeleton_mask -d CTQ.mat -t CTQ.con -f CTQ.fts -n 1000 -C .95 -x -V -D
Loading Data:
Data loaded
Warning: Data demeaning selected, but at least one design column has non-zero mean - therefore invoking automatic demeaning of design matrix
3.95489e+28 permutations required for exhaustive test of f-test 1
Doing 1000 random permutations
Starting permutation 1 (unpermuted data)
[...]
Starting permutation 1000
Critical Value for: CTQ_vox_corrp_fstat1 is: 13.0897
5.09921e+20 permutations required for exhaustive test of t-test 1
Doing 1000 random permutations
Starting permutation 1 (unpermuted data)
[...]
Starting permutation 1000
Critical Value for: CTQ_vox_corrp_tstat1 is: 5.47727
Critical Value for: CTQ_clusterm_corrp_tstat1 is: 6152.31 [<--this also seems high...]
5.60659e+14 permutations required for exhaustive test of t-test 2
[similar for t-tests 2 and 3]
Finished, exiting.
Thanks much,
C.P. Frost
UW-Madison
|