Supplementary material to the paper âAdaptive ... - Etienne Roquain's
BR-2S-α. 1 Robustness when Ï0 and/or Ï are very close to 1. Figure 1 reports a comparison plot between the FDR of the main studied procedures. The case Ï0 ...
Supplementary material to the paper “Adaptive false discovery rate control under independence and dependence” ´ Roquain G. Blanchard and E.
In this note, we present additional simulations carried out in the framework of Section 3.4 of the JMLR paper “Adaptive false discovery rate control under independence and dependence” (we use below the same notation as in the paper). The main point here is to explore additional regions of the parameter space. As reported in Remark 20 of the paper, overall these additional simulations confirm the qualitative conclusions of the paper, in particular the good behavior of the procedure Storey-α and BR-2S-α.
1 Robustness when π0 and/or ρ are very close to 1 Figure 1 reports a comparison plot between the FDR of the main studied procedures. The case π0 = 0.95 is qualitatively similar to the case π0 = 0.8 (reported in the paper in Figure 4, bottom left plot). The case ρ = 0.95 differs slightly from the case ρ = 0.5 (reported in the paper in Figure 2 bottom left plot) to the extent that the FDR of median-LSU (resp. FDR08- 12 ) stays below the target level (resp. can be above the target level, for large π0 ). In both cases, the FDR of Storey-α is still controlled at the target level.
0.02
0.04
0.06
0.08
LSU-Oracle Storey-α Storey-1/2 Median LSU BKY06 BR-1S-α BR-2S-α FDR09-1/2
0
0
0.02
0.04
0.06
0.08
0.1
FDR (ρ = 0.95)
0.1
FDR (π0 = 0.95)
0
0.2
0.4
ρ
0.6
0.8
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
π0
1
Figure 1: Left: π0 = 0.95 , ρ varies. Right: ρ = 0.95 , π0 varies. Other parameters as in the paper (µ = 3 , m = 100 , α = 5%).
1
2 Robustness of Storey-α when both π0 and ρ are ≃ 1
pi0
0.80
0.85
0.90
0.95
1.00
Figure 2 represents the FDR of the Storey-α for (ρ, π0 ) ∈ [0.8, 1]2 . We clearly see that the FDR of Storey-α is below or extremely close to α in that range.
0.80
0.85
0.90
0.95
1.00
correlation
0.035
0.040
0.045
0.050
Figure 2: Top: FDR of Storey-α in function of ρ (X-axis) and π0 (Y-axis). 10000 simulations. α = 0.05. Bottom: the color range.
3 Case of small or large number of hypotheses m The results are qualitatively very similar to the ones shown in Figure 2 of the paper. See Figure 3 on the next page.
4 Case where α = 0.01 and α = 0.1 Fig. 4 shows the behavior of the procedures for other standard values of α . It is qualitatively the same as for α = 0.05.
2
FDR (m = 10, 000) 0.1
0.1
FDR (m = 20)
0.02
0.04
0.06
0.08
LSU-Oracle Storey-α Storey-1/2 Median LSU BKY06 BR-1S-α BR-2S-α FDR09-1/2
0
0
0.02
0.04
0.06
0.08
LSU-Oracle Storey-α Storey-1/2 Median LSU BKY06 BR-1S-α BR-2S-α FDR09-1/2
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
0.08 0.06 0.04 0.02 0 0.1 0.08 0.06 0.04 0.02 0
0
0.02
0.04
0.06
0.08
0.1
0
0.02
0.04
0.06
0.08
0.1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.1
0
π0
π0
Figure 3: Left: m = 20 , right: m = 10, 000; π0 varies; from top to bottom: ρ = {0; 0.5; 0.8} . To reduce the computational burden for m = 10, 000, we performed less repetitions of the experiments, hence the added variability.
3
FDR (α = 1%)
FDR (α = 10%) 0.12 0.1
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
0.8
1
0.12
0
0.005
0.12 0.06
0.08
0.01
0.1
0.015
0.005
0.06
0.08
0.01
0.1
0.015
0.005
0.06
0.08
0.01
0.015
LSU-Oracle Storey-α Storey-1/2 Median LSU BKY06 BR-1S-α BR-2S-α FDR09-1/2
ρ
ρ
Figure 4: Left: α = 1% , right: α = 10%; ρ varies; from top to bottom: π0 = {0.6; 0.8; 0.9} . The latter choice of parameters for π0 was motivated by trying to concentrate on the ranges where the FDR control appeared to be the more likely to be violated from the experiments with α = 5% .
disturbances are homoscedastic with the same variance across time and ... as a Chi Square under the null of homoscedasticity. ..... (b-B) sqrt(diag(V_b-V_B)).
In order to probe the depth of the air cavity between the membrane and the substrate, we performed .... An improved step, beyond the scope of this paper, would.
conductive grain of the order of 250 V/m, while the electric field at a 0.1 ... total kinetic impulse P given by expression (3), as each Universon has an Eu / c momentum : ..... The whole elementary particle of matter is supposed accelerated by an ..
2006), connectance (e.g. the proportion of links established at the scale of the ... resource input will ultimately decrease host population density (thus preventing.
School of Biological Sciences, University of Manchester, Manchester, UK. Running title: Detection and quantification of positively selected sites in HIV.
of the. PCR. Product). He. Ho. Reference. (TC)n repeat primers: TGAAATTCTTCATCAGT ...... Dalebout ML, Ruzzante DE, Whitehead H, Oien NI. Nuclear and ...
The white arrow indicates the puncture hole (Photo: H. Alinezhad). Figure S2. Partially eaten carcass of a Maral Red Deer near the young male leopard probably.
The available scientific evidence does not show that any health problems are ... Whereas high levels of RF can produce health effects (by heating tissue), ...
(ha). Number of trees (first census). Species (% of stems at first census). First ..... Density. 0.000. 0.002. 0.004. 0.006. 0.008. 0. 100. 300. 500. 700. µj densities. µj.
Vercken et alâ by Cote, J., Le Galliard, J.-F., Rossi, J.-M., and Fitze, P. S.. 1. Supplementary ... ventral coloration in female common lizards (n = 86). Scale type.
Mar 22, 2010 - The switch position is controlled by a pulse ... the carrier frequency is to one resonance of the temperature sensor, the stronger ... longer interrogation time: we attribute this slow divergence of the Allan deviation to the simple.
Dec 7, 2012 - observed non-exponential decay for the larger membranes. ... condition can be derived to have the following form,. (S.4) ... features of acoustic transport and solid state physics in general, it is invalid for real materials at.
glacierized area except for a few glaciers that surged and advanced. ... wide flux gate (draining 130 km² of glacier) located just upstream of the area where the.
run a batch algorithm repeatedly to account for new data or model variation, and in ..... Before derivation of these update rules, the first step consists of the estimation of and ..... from the University of Maryland at College Park. (UMCP), in 1988
orders, and compare it to the case in which they can also submit limit orders ( ..... To understand the convergence results of this section, it useful to note that, ...... we obtain a lower bound by ignoring terms n = k, and then multiplying by the .
Oct 9, 2008 - to having a key function in cognitive control. ... supplementary motor area (pre-SMA) are, in humans, .... that the electrical excitability of the SMA is greater ...... there is an essentially autonomous supervisory (executive) system t
This table also gives research guidelines, since all this ...... Agave studies in Yucatan, Mexico. ... The Domestication of Pineapple: Context and Hypotheses.
There was, however, a significant difference be- tween the sex groups, F (1, 60) =7.69, p < .01, with males responding significantly faster than females (377 vs.
is the engine of the modelâ .... firms in the oil and energy sector, the wild swings in world-wide energy prices make too poor a proxy of total factor productivity.