Throughout the simulations presented above, we have adjusted only for those axes of genetic variation amongst the first three, which are correlated with disease status using P <0.05. For example, if only the first and third of these axes are correlated with disease status, we do not make adjustment for the second. We have investigated the sensitivity of our approach to this selection strategy by repeating simulations: (i) adjusting only for those axes of genetic variation amongst the first three that are correlated with disease status using P <0.1 and (ii) all of the first three axes of genetic variation, irrespective of their correlation with disease status.
