TABLE 1.

Power to detect main-effect and interactive loci in a SNP association study using selective genotyping in the context of a three-QTL model, using a selection fraction (α = 20%), and treating the selectively genotyped sample as a case–control study (see text for details)

		Power (%)
Three-QTL model				Single-term model
Parameters	β1 (% QTL variance)	β1	β2	β3	β12	β1	β2	β3	β12
r = 0.5; n = 800	0.1 (0.5)	96	55	78	78	96	52	79	80
	0.2 (2.0)	100	53	75	75	100	56	77	79
	0.4 (8.0)	100	50	72	66	100	53	80	78
	0.6 (18.0)	100	43	67	49	100	51	78	80
	0.8 (32.0)	100	37	58	34	100	52	77	82
r = 0.8; n = 2500	0.1 (0.3)	96	51	96	85	97	50	96	88
	0.2 (1.3)	100	48	97	82	100	54	97	88
	0.4 (5.1)	100	42	93	74	100	55	97	86
	0.6 (11.5)	100	33	92	62	100	54	97	86
	0.8 (20.5)	100	28	87	57	100	56	96	87

The power corresponds to tests with a significance level of 0.10. It was estimated using 10,000 simulations, and the estimates are correct to ±1%. In the simulations, β₀ = 0, β₂ = 0.05, β₃ = 0.07, and β₁₂ = 0.1; n is the sample size of the “case–control” sample; the actual sample size is 5n. We denote by r the common major allele frequency for the loci. The power to detect the main-effect β₁ increases with β₁, as expected, but the power to detect other loci and the interaction term (β₂, β₃, and β₁₂) decreases with β₁. The decrease in power is most severe when the variance explained by the strongest locus exceeds 10%.