24.02.2013 Views

Optimality

Optimality

Optimality

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

0.025<br />

0.02<br />

0.015<br />

0.01<br />

0.005<br />

On the false discovery proportion 45<br />

α i ’’<br />

0<br />

0 10 20 30 40 50<br />

i<br />

60 70 80 90 100<br />

η<br />

x 10−3<br />

4<br />

3<br />

2<br />

1<br />

i ’<br />

0<br />

0 10 20 30 40 50<br />

i<br />

60 70 80 90 100<br />

8<br />

6<br />

4<br />

2<br />

α i ’’/η i ’<br />

0<br />

0 10 20 30 40 50<br />

i<br />

60 70 80 90 100<br />

Fig 1. Stepdown Constants for s = 100, γ = .1, and α = .05.<br />

that of independence of p-values used by Benjamini and Liu [2].<br />

Theorem 4.1. For testing Hi : P ∈ ωi, i = 1, . . . , s, suppose ˆpi satisfies (2.1).<br />

Consider the stepdown procedure with constants<br />

(4.1) α ∗ i = min{<br />

sα<br />

,1}<br />

(s−i+1) 2<br />

and assume the condition (3.5). Then, FDR≤α.<br />

Proof. First note that if|I| = 0, then FDR = 0. Second, if|I| = s, then FDR =<br />

P{ˆp (1)≤ α ∗ 1}≤ � s<br />

i=1 P{ˆpi≤ α ∗ 1}≤sα ∗ 1 = α.<br />

Now suppose that 0 α ∗ 1). Define t to<br />

be the total number of true hypotheses rejected by the stepdown procedure and<br />

f to be the total number of false hypotheses rejected by the stepdown procedure.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!