Affymetrix), amounting to 0.53 and 0.63 % of imputed SNPs (5,000–6,000 false positives per one million imputed SNPs) in European Americans and African Americans, respectively. These results are consistent with Sinnott and Kraft, who estimated an average false positive rate (based on the genome-wide significance threshold of P < 5 × 10−8) of 0.4 % among 2,347,809 imputed SNPs based on a study sample composed of healthy control groups of European descent who were genotyped on Affymetrix 6.0 (subjects arbitrarily designed as cases) or Illumina 550v1 (subjects arbitrarily designated as controls) (Sinnott and Kraft 2012). They observed false positive rates as high as 1.3 % when imputing SNPs genotyped from Illumina but not Affymetrix. Similarly, Uh et al. (2012) reported a genomic control inflation factor well above 1.0 (λgc = 1.16), indicative of many false positive associations when imputing across Affymetrix and Illumina arrays. In the current study, there was no evidence of false positive associations among the genotyped SNPs for any pair-wise set of arrays, strongly suggesting that the observed bias among the imputed SNPs is due to the imputation process rather than differences in genotyping quality. Moreover, it is clear from these studies that the degree of bias
