In this example we analyze the full POPRES sample. Smartpca’s six standard deviation outlier rule removes 141 outliers, including all of the E. Asian and Mexican samples. If these “outliers” were retained, PCA finds 12 significant dimensions: the first four dimensions separate the five continental populations (African, European, Latin American, E. Asian, and S. Asian). Other ancestry vectors are difficult to interpret. Moreover, based on this embedding, Ward’s clustering algorithm failed to converge; thus no sensible clustering by ancestry could be obtained.
