Both IQS and imputation accuracy compare true genotypes to imputed genotypes. Given that imputation is designed to infer unknown genotypes, one purpose of this paper was to use IQS to evaluate statistics that measure the quality of imputation without knowing the true genotype. The two statistics most commonly used for this purpose are the variance ratio (rsq_hat in MACH)[10] and the imputed information score (PROPER_INFO in SNPTEST) [5]. The variance ratio for a particular SNP is a ratio of the empirically observed variance (based on the imputation) to the expected binomial variance p(1-p), where p is the minor allele frequency[18]. As the amount of information available to impute the SNP decreases, the empirically observed variance decreases and the variance ratio approaches zero. The product of the variance ratio and sample size defines the ‘effective sample size’. Similarly, the imputed information score is a measure of genotype information content, which is related to the effective sample size (power) for the genetic effect being estimated [1], [5], [18]. Although computed using a different approach, the information score is analogous to the variance
