The effective application of our method depends on an appropriate estimate of the variance parameters. The IBS or Balding-Nichols matrix43 appears to be better than IBD estimates at capturing the long-distance relationships that result in variations at the population level. However, when the structure of the sample at hand is better described in terms of fairly recent hidden relatedness, methods based on the estimation of IBD may have an advantage. In principle, our approach is also suitable for association testing in a data set including individuals from a heterogeneous population with admixed background. In such cases, it is important to consider SNP ascertainment bias in estimating the degree of relatedness between individuals. Because many SNP probes in genotyping arrays are selected from European populations, the marker-based pairwise distance between two individuals may appear to be larger between unrelated European samples than between unrelated individuals from other populations. To resolve the resulting ascertainment bias, each SNP may be differently weighted when the IBS similarity matrix is computed. A general framework has been presented19 for computing the similarity matrix with a different
