λ=mediani=1,…,m{Xi}/0.455 where 0.455 is approximately the theoretical median of a χ12 distribution (Devlin and Roeder 1999). Our model assumes the null distribution of each Xi is χ12, so if this assumption is valid, we should have λ ≈ 1. A value of λ > 1 suggests that the observed variance of the test statistic is larger than the theoretical variance, which will tend to increase the number of false positives. We also calculated the percentage of SNPs significant at the 5 × 10−8 significance level, a standard significance level used for GWA studies (McCarthy et al. 2008). Assuming the genotype is measured accurately, we don’t expect genotype frequency differences between our cases and controls, because they are both samples of healthy women used as control groups for other studies. Thus, we should see very few SNPs with such significant p-values (approximately 1 out of every 20,000,000 independent tests).
