To perform assignment, we first estimated the β coefficients by means of least-squares regression with a training set of individuals with known locations and then used the estimated coefficients of the linear model to predict the latitude and longitude of a test individual on the basis of their PC1 and PC2 values (we call this a ‘continuous assignment’). We also made a ‘discrete assignment’ by assigning individuals to the country for which the centre-point is closest to the latitude and longitude predicted by the continuous assignment method. In practice, the two methods produce roughly similar results (Supplementary Table 4). As a reference point for evaluating performance, the Supplementary Table also reports statistics for how a method would perform if all individuals were assigned to a central location within Europe (here taken to be Austria).
