To ameliorate the observed bias, Sinnott and Kraft (2012) tested three methods of correction: (1) use of principal components as covariates in logistic regression analyses; (2) restricting analyses to imputed SNPs with high accuracy, up to R2 = 0.99; and (3) genotyping a subset of controls on the array used for cases to screen out problematic SNPs. Only genotyping a subset of controls provided a level of correction that would avoid a large number of false positive associations (Sinnott and Kraft 2012). However, this correction method is not applicable for use in studies without access to original study DNA or where budgets would not allow for additional genotyping; both are important limitations when using publicly available genotype data. Uh et al. (2012) proposed a strategy of post-imputation filtering using their RT2 statistic (RT2 ≥ 0.98), which was analogous to Sinnott and Kraft’s filtering on R2 = 0.99 but for a sibling pair plus control design. Both post-imputation filtering strategies substantially reduced the observed bias in the SNPs meeting their filter requirements, but unacceptable error rates remained (500 false positives for
