A standard test for association in GWAS is the Armitage trend test21,29, equal to N times the squared correlation between genotypes G (0, 1 or 2) and phenotypes Φ (0 or 1), where N is the number of samples. This statistic extends to imputed data by using genotype dosages. The value of the statistic decreases by a factor of r2 if computed at a genotyped or imputed SNP in partial LD with the causal SNP13. To estimate the expected association statistic in a GWAS over a set of N samples sequenced at average coverage c, we first estimate the accuracy r2(c) attained at coverage c by sub-sampling IHCS data. We then estimate the expected association statistic as Nρ2(G, Φ) r2(c).
