For each of the 21,800 probes (18,226 unique autosomal genes) selected for analysis, we performed a cis- association test between expression and common SNP genotypes using a Spearman Rank Correlation (SRC) model (See Methods). The model was applied to each population separately: 1) For the normalized and stratification-corrected expression data, and 2) for the REDUCED expression data. The purpose of presenting both sets of results is to compare the properties of the two sets, as opposed to simply choosing one approach over the other. We analyzed in depth those associations significant at the 0.01 permutation threshold. At this level of significance, we expect roughly 182 genes to have at least one significant association by chance, and we detected 657, 774, 698, 795, 773, 472, 947, 799 genes with a significant association in CEU, CHB, GIH, JPT, LWK, MEX, MKK, and YRI, respectively with a false discovery rate (FDR) of 18–39% per population (Table 1). Similar FDR values were obtained when evaluating the degree of replication of an eQTL discovered in one population by replication in another (see Methods) where FDR
