I ran multiple comparison correction on my data using both FWE and FDR. The most significant spots
on the snpm T map (snpmT-.img) were represented with the FWE correction (snpmt_filtered.img), but
the filtered FDR map using a .05 threshold included voxels with very low t values (<1) and even
excluded some with very high t values. These spots with low t values are clearly not significant, so
why were they represented on the filtered map? The FDR procedure spit out an uncorrected p value
of .004 as being necessary for significance. Since the filtered figure does not appear accurate, would
it be valid to find the t value associated with p=.004, df=12 (probably around t=3.5) and threshold
my t image to this value for displaying significant results of the FDR procedure?
|