Sponsored by:
Assortative mating (AM) is known to influence genetic studies, an issue that has been receiving increasing attention in recent years. We extend the existing theory for more general models of sorting, concluding that theory-based AM corrections require knowledge of historical sorting patterns that are complicated and unknown. We propose simple approach using polygenic indexes that can estimate the fraction of genetic variance and genetic correlation is driven by AM without additional information. The same approach applied to Mendelian randomization (MR) studies can also correct for bias due to long-range linkage disequilibrium arising from AM. We show through theory and simulation, however, that AM induces an additional form of selection bias in MR studies under certain circumstances that remains after our adjustment, and we derive a necessary condition for the presence of such bias. Using data from the UK Biobank, we apply our methods to studies of education and several health measures. We find that AM inflates genetic correlation estimates between education and health by an average of 14 on average. Our results suggest a cautious approach is needed in interpreting genetic correlations or MR estimates, especially when one or both traits is subject to assortative mating.
Marta Bilghese, University of Southern California
Paul Auer, Medical College of Wisconsin
Dan Benjamin, University of Southern California
Michael Livermore, University of Southern California
Miles Kimball, University of Colorado Boulder
Patrick Turley, University of Southern California