Hi Yi-Yu,
I'd say that "more suitable" might be a too imprecise term. More powerful?
More conservative? More liberal? The "suitability" might depend on the
particular application. Both pFDR and FDR have their pitfalls.
The Storey's paper "Direct approaches to false discovery rates" (2002)
states that pFDR would make more sense because when the rejection region
converges to zero, pFDR converges to a value close to the proportion of true
null hypotheses, while FDR also converges to zero. However, if all
hypotheses are null, while pFDR converges to 1 (see fig 4 on this paper),
FDR controls over FWE in the weak sense, which might also be seen as "more
suitable" in some settings.
If you are planning to make a statement like this in a paper, and the
reviewer is asking you to explain it better, I think the right thing to do
would be to explain why in your particular study pFDR is expected perform
"better" than other multiple testing procedures.
Hope this helps!
Best regards,
Anderson
|