Hi Ajay,
> I had a question in regards to
> family-wise error vs FDR correction.
> Which of these would show more results when doing a two
> group comparison?
> Is one more stringent than the other?
This one is for Tom or Steve... I guess it depends on your data and on what you control for fwe (voxel, cluster, "cluster enhanced" voxel).
> I tried to use
> FDR after running
> randomise but I received statistical significance in
> several areas across
> the brain as opposed to the corrp_tstat1 file(familywise
> error). I am not
> sure if my FDR is correct or which of the two results is
> more valid?
If you had significant results with one or the other, both would be valid.
> Below are the fdr commands I used.
>
> randomise -i GM_mod_merg_s2 -m GM_mask -o GM_mod_merg_s2 -d
> design.mat -t
> design.con -T -n 5000 -V
> >>output file is GM_mod_merg_s2_tfce_p_tstat1
>
> fslmaths GM_mod_merg_s2_tfce_p_tstat1 -mul -1 -add 1
> 1_minus_GM_mod_merg_s2_tfce_p_tstat1
> fdr -i 1_minus_GM_mod_merg_s2_tfce_p_tstat1 -m GM_mask -q
> 0.05
> >>after this step it said the threshold is 0 even
> though I entered 0.05, so
> not sure why?
You're doing the right thing, but I'm afraid this 0 means that your results do not survive the FDR correction.
> fslmaths GM_mod_merg_s2_tfce_p_tstat1 -thr .95
> thresh_GM_mod_merg_s2_tfce_p_tstat1
> fslview $FSLDIR/data/standard/MNI152_T1_2mm
> thresh_fslvbm_tfce_p_tstat1 -l
> Red-Yellow -b 0.95,1
These are your uncorrected p-values. As you mentioned earlier on, you would need to check if you've got anything significant in any of your corrp maps (tfce, clustere and/or clusterm).
Cheers,
Gwenaëlle
