Three data sets were compared for each pair of studies: (1) genotyped SNPs shared on both arrays; (2) imputed SNPs based on the union of genotyped SNPs available on either array; and (3) imputed SNPs based on the intersection of genotyped SNPs available on both arrays. The first analysis tested for any potential genotyping bias that might affect imputation results. The second and third analyses were designed to test the magnitude of bias resulting from imputing the same SNPs based on either the union of genotypes SNPs (which uses the maximal information available) or the intersection of genotyped SNPs across arrays (which corresponds to less input information). Statistically significant SNP associations were identified as those having P < 1 × 10−6, based on Bonferroni correction for the largest number of SNPs in any one of our analyses (N = 43,035 SNPs). Since case or control status was arbitrarily assigned, inflated λgc values and significant associations between SNPs and case status demonstrate systematic imputation bias as evidenced by false positive or spurious associations.
