Sponsored by:
There is a strong documented relationship between education and health, both phenotypically and genetically. This has given rise to an expansive literature reporting genetic correlations (rg) and Mendelian Randomization (MR) estimates trying to identify and leverage common genetic factors between them. However, some of this relationship is due to assortative mating (AM). We develop a method to estimate the component of rg that is not driven by AM and we apply it to data on educational attainment and eight measures of health in the UK Biobank. Our method to detect and adjust for AM can be employed in samples of unrelated individuals (without data on mate pairs) by exploiting the effect of long-range linkage disequilibrium on polygenic indexes. We find that our AM-adjusted estimates of rg between education and health are moderately smaller than our raw rg estimates, ranging from a 6.2 reduction (cigarettes per day) relative to the unadjusted rg. Finally, we propose a related method to correct MR estimates for AM. Preliminary simulation results imply that this approach may be valuable in the future, when larger samples are available, but that current sample sizes are too small to precisely correct MR estimates using PGIs and long-range LD.