Large, genetically heterogeneous data sets are routinely analyzed for genome-wide association studies. These samples exhibit complex structure that can lead to spurious associations if differential ancestry is not modeled. Numerous approaches for handling this issue are now available in the literature [Epstein et al., 2007; Pritchard et al., 2000; Purcell et al., 2007; Zhang et al., 2003]. Due to computational challenges encountered in genome-wide association studies much interest has focused on estimating ancestry using computationally efficient methods such as PCA. These methods are based on an eigenvector decomposition of a matrix that reflects genetic similarity between pairs of individuals.
