Table 1.

Type I error of GHC is estimated in each setting with 50 million simulations. The strong LD setting is based on a subset of eight FGFR2 HapMap SNPs that are in high LD with one another. The weak LD setting is based on a subset of eight FGFR2 HapMap SNPs that are in weak LD with one another. For reference, we also include the Type I error of the original higher criticism ignoring the presence of correlation, with p-values computed as in Barnett and Lin (2014).

Significance Level	Strong LD (FGFR2 subset)	Weak LD (FGFR2 subset)	Random chr5 genes	Random chr5 genes (original higher criticism)
α = 5.0 · 10−2	4.62 · 10−2	4.91 · 10−2	4.64 · 10−2	7.23 · 10−2
α = 1.0 · 10−2	9.59 · 10−3	1.04 · 10−2	9.53 · 10−3	1.38 · 10−2
α = 1.0 · 10−3	9.63 · 10−4	1.08 · 10−3	9.80 · 10−4	1.19 · 10−3
α = 1.0 · 10−4	9.51 · 10−5	1.00 · 10−4	9.63 · 10−5	1.26 · 10−4
α = 1.0 · 10−5	9.70 · 10−6	9.50 · 10−6	8.05 · 10−6	1.26 · 10−5
α = 1.0 · 10−6	8.60 · 10−7	7.00 · 10−7	8.63 · 10−7	1.30 · 10−6