When missing genotypes are imputed by MaCH, each SNP has an R2 value associated with it that quantifies the quality of the imputation. The R2 value is an estimate of the squared correlation between the imputed genotype and the actual genotype, so a higher R2 corresponds to a SNP imputed with more certainty. Standard advice is to restrict to SNPs with R2 > 0.3, which we did (Scott et al. 2007). It is expected that this will remove 70% of poorly imputed SNPs while keeping 99.5% of better imputed SNPs (Li et al. 2010). To reduce the error inflation in our less standard setting, we considered restricting to SNPs imputed at even higher quality.
