Table 2.

Impact of G-STRATEGY on Type 1 Error for Various Association Tests

			Empirical Type 1 Error (SE) with Nominal Type 1 Error of
			.05		.01
Association Test	Trait Model	MAF	n0 = 0	n0 = 100	n0 = 0	n0 = 100
MQLS	1a	.4	.046(.003)	.053(.003)	.008(.001)	.007(.001)
MQLS	1a	.2	.053(.003)	.048(.003)	.011(.002)	.008(.001)
MQLS	1a	.05	.054(.003)	.051(.003)	.010(.001)	.011(.001)
MASTOR	4a	.4	.051(.003)	.050(.003)	.009(.001)	.008(.001)
MASTOR	4a	.2	.051(.003)	.053(.003)	.009(.001)	.011(.001)
MASTOR	4a	.05	.050(.003)	.048(.003)	.009(.001)	.010(.001)
GTAM	4a	.4	.049(.003)	.052(.003)	.011(.002)	.010(.001)
GTAM	4a	.2	.049(.003)	.049(.003)	.009(.001)	.009(.001)
GTAM	4a	.05	.051(.003)	.047(.003)	.008(.001)	.009(.001)

Note: In each case, the empirical type 1 error is based on 5,000 simulated replicates. None of the empirical type 1 error values is significantly different (p < .05) from the nominal level based on a z-test. Among the 960 sampled individuals, n₀ = 0 or 100 randomly selected individuals are assumed to be previously genotyped. In each case, we implement G-STRATEGY to select 300 − n₀ additional individuals to be genotyped, so as to obtain a total of 300 genotyped individuals. Association is tested on an unlinked, unassociated, binary variant using various test statistics.