Table 1. Power of different methods, on 1000 data sets and a region with 100 rare SNVs.

Error rate	Coverage per person	C-alpha	Freq-weight	VT	RWAS	LRT_G	VST	LRT
0	4	0.062	0.659	0.872	0.668	0.817	0.973	0.975
	10	0.062	0.713	0.903	0.688	0.854	0.98	0.99
	20	0.067	0.722	0.903	0.692	0.858	0.987	0.988
0.01	4	0.051	0.174	0.208	0.269	0.286	0.634	0.745
	10	0.066	0.479	0.674	0.610	0.71	0.965	0.972
	20	0.067	0.686	0.887	0.690	0.848	0.988	0.993

Two error rates (0% and 1%) were considered with three different coverages (4, 10, and 20). Population attributable risk of the region was 0.02, with c_i = 0.1 and a significance threshold of 0.05. Tested methods were C-alpha (Neale et al. 2011), Freq-weight similar to the Madsen–Browning method (Madsen and Browning 2009), variable threshold (VT) (Price et al. 2010), RWAS (Sul et al. 2011b), LRT_G (Sul et al. 2011a), and our proposed methods (VST and LRT).