If you want to see the FDR q-value in the AFNI GUI, run the command
3drefit -addFDR ANOVAdataset+tlrc
If you need to use a mask, add the option
-FDRmask maskdataset+tlrc
before the input dataset name.
See the output of '3drefit -help' for a few more details. For lots
more information on FDR in AFNI, see slides 44-53 of the presentation
below:
[
afni.nimh.nih.gov]
[
afni.nimh.nih.gov]
The '3drefit -addFDR' command will compute a curve giving q as a
function of F for each F statistic sub-brick, and then as you change
the F slider threshold, the corresponding q-value is displayed at the
bottom of the slider. In addition, if you hover the mouse cursor over
AFNI's p=/q= label you will see a 'popup hint' that shows the MDF = an
estimate of the 'missed detection fraction' = fraction of true
positives that lie BELOW the current threshold. (MDF=60% is NOT saying
that 60% of the voxels below threshold are true positives; it is
estimating that only 40% of the true positives are above threshold and
the rest are below threshold.)
MDF is sort of the converse to FDR = fraction of positives ABOVE the
current threshold that are false. Estimating MDF reliably is much
harder than FDR, so I wouldn't put a huge amount of reliance on it --
but it can be instructive, since at usual thresholds, MDF is roughly
50-70%, indicating that we might be missing a lot of stuff with our
current FMRI analyses .
3dFDR is less useful -- it is an older program. It replaces the F-
statistic volume with the equivalent z(q)-values. You CAN threshold
with this, but I much prefer thresholding with F (or t) and just
seeing the q-value below the slider.