P-value and FDR pipeline

AFNI Message Board
1

Hi everyone,

We have a pipeline that we’re very familiar with for processing functional maps that gives t-stat maps thresholded to the top 97, 98 and 99th percentile of voxels. We used 3Ddeconvolve to generate a bucket with one sub-brick for the stimulus. However, we want to compare to see the results with thresholding at p-values with FDR correction (ex. p <0.001 with FDR). Is there a recommended or standard pipeline that anyone can recommend or suggest?

Cheers,
Christopher

2

we want to compare to see the results with thresholding at p-values with FDR correction (ex. p <0.001
with FDR). Is there a recommended or standard pipeline that anyone can recommend or suggest?

FWE and FDR are two totally different approaches to handling the multiple testing problem. In other words, they deal with the multiple testing issue from two different perspectives, and they usually don’t fully agree with each other in terms of the final results.

3

I think I might be misunderstanding the role of FDR in thresholding - after reading some more, am I correct in that there is no “FDR-corrected P-value” per se, but rather, you pick the P-value based on the choice of Q in FDR? I.e. there is no “FDR-corrected P < 0.001 threshold” but a threshold of a chosen Q-value (say, 0.05) that has an accompanying P-value?

Cheers,
Christopher

4

am I correct in that there is no “FDR-corrected P-value” per se, but rather, you pick the P-value based
on the choice of Q in FDR? I.e. there is no “FDR-corrected P < 0.001 threshold” but a threshold of a
chosen Q-value (say, 0.05) that has an accompanying P-value?

Yes, that’s correct.