BH, why this threshold ?

[NN_STAT]

Hello,

I have a theoretic question. When using the BH procedure for controlling the false discovery rate, we compare the p values in this way:

P(i)<=t*(i/m)

where i is the test sorted by the p values, and m is the number of hypotheses done.

I don't understand the logic behind this size that we compare the p values to. For example, in Bonferonni, the logic is clear. If you have 10 tests, divide rejection threshold by 10, it's very intuitive. In BH's case, I don't get the intuition behind it. Can anyone explain it in a simple way ?

Thanks !

[Dason]

Have you read the original paper? I believe there is some intuition given in the paper. I could probably give some more details if you have a hard time getting the intuition from the paper but at the moment I'm too lazy (hey - at least I'm honest).