20 Oct
2005
20 Oct
'05
4:22 p.m.
Hi,
We've been having alot of discussion in our group regarding how one should set significance levels. Since you include FDR in the newer releases, I take it that is your current reccomendation (see also Doug's slides: http://surfer.nmr.mgh.harvard.edu/docs/ftp/pub/docs/fsfsl-mris_glm.pdf). Correct?
Put very roughly, the FDR rate controls the (expected) percent of your positive results that are false. So what level is appropriate? All that Genovese says is that "it is common to set q to conventional levels ...(e.g. 0.01-0.05) ... values in the range of 0.10-0.20 are reasonable in many problems (Benjamini, personal communication)" [quote from pg 872, 2nd column].
What sort of rates do you think are reasonable, and under what conditions?
+glenn