For example, let's say that in the B-H formula [ p < (i/m)q ] q is set to .15 to create the critical value. Would I report that my (FDR) q-value is < .15?
Also, the result of the equation above would be called the "critical value" but the B-H adjusted p-value is different from the critical value. Is that correct?
Take a look at the attached paper on B-H. I think it will help. Best, D. Booth
I found out more about the B-H procedure and here is how I understand it at this point. This is a real email to a colleague but I changed some info to disguise the data:
There were only 8 comparisons (8 select variables frequently tied to Parkinson's) all in relation to EC. I calculated these values using the B-H equation;
(I/m)Q = the critical value I= the individual p-value's rank m= the total number of tests Q= the false discover rate (chosen by the researcher and in our case .15)
I was a little confused for a while about how to interpret the Q, critical, and p-values in relation to each other so I checked in with some stats discussion forums and also read through the original B-H paper myself. Essentially, what I learned was that the highest p-value that is also smaller than the critical value is the cut off point and everything above that is significant. I decided not to report B-H adjusted p-values because, from what I read, they were somewhat misleading and often interpreted incorrectly.
- Badges
- Science topic
- Similar topics
- Correction