First, for the HPFS project, we estimated the concordance between a HapMap control (NA12144) run on each plate and the consensus genotype calls of 139 replicate genotyping instances of this subject from an independent study using the same array and calling algorithm (and using SNPs that pass the quality filters described below). The mean discordance is very low for plates with and without low quality samples and the difference is not significant (15.8 × 10−5 for 15 plates with low quality samples and 9.8 × 10−5 for 14 plates without such samples; p-value=0.72). Second, we re-called a sample of 8 plates from each study, four plates with and four plates without low quality samples. The maximum number of low quality samples per plate was five. The discordance between the original and recalled genotypes is significantly higher for plates from which low quality samples were removed (p-value=9 × 10−4) and it varies significantly with the number of samples removed (p-value=2 × 10−4). However, the discordance is very low for both types of plates. The highest median discordance between original and recalled
