You appear to be using differing thresholds for your corrected vs.
uncorrected comparison. Why would it be surprising to find that an
uncorrected map thresholded at 1-P = 0.995 looks similar to the
corrected map at 1-P = 0.95? You're finding that using a more
"stringent" threshold for the uncorrected map appears similar to a less
stringent threshold in the corrected map, which is exactly what you
On Tue, 2010-06-22 at 20:48 +0100, Jared Moreines wrote:
> Hello all,
> I've been having a rather confusing result from a TFCE randomise TBSS analysis of FA differences between two groups. Whereas previous posts have reported smaller regions of findings in the corrected as compared to the uncorrected p-value images, I seem to be experiencing large between-group differences in FA that are significant in the tfce_corrp_tstat2 image (min 1-P threshold .95) but not in the tfce_p_tstat2 image (min 1-P threshold .999). However, if I lower the uncorrected image threshold to a more liberal .995, it begins to look more like the corrected image.
> I know it is common for findings that are significant in the uncorrected image to not be significant in the corrected image, but this seems to be the opposite scenario. My question is how could there be such a significant group difference in the corrected p-value image that is not reflected in the uncorrected p-value image? This seems statistically counter-intuitive. Does it have something to do with tfce being more sensitive to small differences in FA that are present in large clusters?
> All analyses were performed using 10,000 permutations:
> randomise -i all_FA_skeletonised -o PT_vs_CONTR_T2_n10000 -m mean_FA_skeleton_mask -d PT_vs_CONTR.mat -t PT_vs_CONTR.con -n 10000 --T2