For various thresholds (α = 0.001,0.01,0.1,0.2) and various numbers of additional controls on Affy (n = 100,300,500) we removed SNPs significant at level α in a preliminary analysis comparing the n additional Affy controls to the 1000 original Illumina controls. We then performed standard logistic regressions comparing the 1000 Illumina controls to the 1000 Affy cases using the remaining SNPs. The genomic control λ and the percentage of highly significant SNPs were calculated; results for the soft call genotypes are shown in Table 2. As n increased and α increased, our summary measures improved. This again happened at the expense of losing SNPs for analysis, but not as quickly as in Method 2. Figure 4 shows the discriminatory ability of this method for each n as we vary the α screening threshold.
