Unlike genomic control, the EMMAX model alters the ranking of SNPs by their association statistics. This is especially important as many recent GWAS follow-up and multistage design studies take the approach of genotyping all SNPs exceeding some predefined threshold29-31. We examined the extent to which the adoption of the EMMAX model changes the SNP rankings in comparison to the uncorrected and principal component analyses. We took the top k markers from the results of EMMAX, the uncorrected method, and regression including 100 principal components (as implemented in EIGENSOFT software), for k between 10 and 5,000. For each of these sets, we calculated the number of SNPs shared between the lists and the fraction of these shared SNPs relative to the number of unique SNPs in each pair of lists. Although many of the top SNPs reported by each method overlap, a considerable number of highly ranked SNPs differ between the methods (Fig. 4 and Supplementary Table 2). In general, EMMAX results are similar to uncorrected analysis when the inflation of test statistics is small, but they become more similar to
