We also applied our method to the WTCCC data set consisting of case-control studies for seven common diseases6. To analyze case-control phenotypes, we applied a linear model to the binary phenotypes, in the spirit of Armitage’s test (see Online Methods). We performed association testing over the seven disease phenotypes using EMMAX, EIGENSTRAT and uncorrected analysis. The values we observed for inflation factors λ were very similar to those in the original study, in which the test statistics were uncorrected: bipolar disease, 1.11; coronary artery disease, 1.06; Crohn’s disease, 1.10; hypertension, 1.06; rheumatoid arthritis, 1.03; type 1 diabetes, 1.04; and type 2 diabetes, 1.07. Consistent with our observations over the NFBC66 data, correcting for 100 principal components only partially reduced the inflation factors (Table 3 and Supplementary Fig. 2). When EMMAX was applied, the estimated inflation factors were below the upper bound of the confidence interval, suggesting that none of the phenotypes show significant inflation of test statistics.
