For each replicate of data, we apply classical multidimensional scaling to the IBS relationships between each pair of samples. Given that we simulate four populations (the original source population and three external cohorts), we would expect to require up to three axes of genetic variation to distinguish between them. We thus utilize the forward-selection procedure, described above, to determine which of the first three axes of genetic variation are correlated with phenotype. Three likelihood ratio trend tests of association are then performed:
