See all my videos at www.tilestats.com/ 1. FDR and BH method 2. Storey's method (01:20) 3. Estimate the FDR (02:38) 4. q values (09:22) 5. Storey vs BH vs Bonferroni (13:48)
The BH has been around longer and is easier to implement. Also, Storey's method requires that you have a very large number of p-values so that you can analyze the distribution of these. That might be a few reasons.
Thanks for the informative video. But be careful - Storey's method can break down when a large proportion of the null hypotheses are false, see "Paradoxical results of adaptive false discovery rate procedures in neuroimaging studies", Neuroimage. 2012 Dec;63(4):1833-40. doi: 10.1016/j.neuroimage.2012.07.040. Epub 2012 Jul 27.
Thanks for sharing your knowledge. I wish I had included that Storey's method might reject more H0 than if one uses the unadjusted p-values, if a large proportion of the H0 are false, which is the nature of the FDR.
The only use of the first part of the calculation (8:48) was to get the m0 from lambda? in other words, it is saying p=0.05 is q=0.0739. The second part (10:37) uses m0 to calculate all the q-values from largest to smallest. I wonder if one can calculate specific q-values, just using the first part of the math. Say p= something, I should get the q for that p. why use the part 2 math to calculate when part 1 is working great?
