Dear all,
I am currently using the function nilearn.glm.fdr_threshold from nilearn and encountered an unexpected behavior.
When applying the function to a vector of z-values, I obtain an infinite threshold. However, when I apply it to the negated version of the same z-values, I obtain a finite threshold (around 3.1).
This makes me wonder whether the function is effectively computing the FDR threshold using only positive z-values.
Could anyone clarify how fdr_threshold handles positive vs negative statistics?
Thank you very much for your help!
Best regards,
Manon Pietrantoni