We randomly selected 1000 subjects from the 1,038 on Illumina to serve as controls, and 1000 subjects from the 1,672 on Affy to serve as cases. Then from the remaining 672 subjects on Affy, we selected n additional subjects to serve as controls genotyped alongside cases on the Affy platform. We first performed a screening step, in which we compared these n Affy controls to the 1000 Illumina controls and eliminated SNPs significant at level α. Then, restricting to SNPs that passed this screening, we performed the main analysis, comparing the 1000 Illumina controls to the 1000 Affy cases, and calculated the Genomic control λ and the percentage of SNPs with p < 5 × 10−8 in this main analysis. We did this calculation for n = 100,300 and 500, and for α = 0.001,0.01,0.1, and 0.2. We also constructed ROC curves to assess the discriminatory ability of this method while varying α, the screening threshold. That is, as we varied the α screening threshold between 0 and 1, we compared how many problematic SNPs (in the main analysis of 1000 Illumina controls vs. 1000 Affy cases) were being detected to how many non-problematic SNPs were being excluded.
