The ancestral composition of samples for genome-wide association can be highly variable. Often, however, the majority of the data come from one continent, but a sizable fraction of the sample derives from other continents. To mimic this scenario we created a subsample from POPRES including 832 Europeans (all self-identified British, Italian, Spanish, and Portuguese subjects), 100 African Americans and 100 Asian-Indians. Using smartpca, seven dimensions of ancestry are significant. The first two eigenvectors separate the continental samples. The third and fourth eigenvectors separate the Europeans roughly into three domains (Fig. 3). The three European populations cluster, but they are not completely delineated. The other continental groups generate considerable noise near the center of the plot. The remaining three significant dimensions reveal little structure of interest.
