In the third dataset, which was created and analyzed by Mark Shriver and colleagues [5], we have data from 12 populations. The missing data pattern showed some evidence of population structure, with the missing data concentrated in particular samples, populations, and SNPs. For this reason, we only used markers for analysis for which there was no missing data, and we corrected for LD using our regression technique (see below). The details of the data preprocessing steps are described in Methods. We analyzed samples from 189 individuals on 2,790 SNPs. On this dataset, we find the leading eigenvalue statistics to be as shown in Table 3.
