Estimating Returns to Education Using the Genetic Lottery

Do more years of schooling cause higher lifetime earnings? Social scientists have long sought to determine the economic returns to schooling given its central importance to individual educational choices and public policy. Prior estimates are limited due to unobserved confounding in observational studies and the restriction of popular school-reform-based quasi-experimental estimates to schooling increases at one specific age only. Genotyped data offers a new opportunity to address unobserved confounding and estimate the returns to one general additional year of schooling by using genetic variants related to education, randomly assigned within families at conception, as instrumental variables (Mendelian randomization (MR)). We combine this design with comprehensive Norwegian population-wide registry data and genotyped data from the Norwegian Mother, Father and Child Cohort Study. We employ three identification strategies for triangulation. MR implies returns to schooling of 8.0% (95% CI=7.1-9.0; USD 5,586; N=110,049), with similar magnitude from sibling MR (6.3%, 95% CI=1.5-11.6; N=18,700), and pleiotropy-robust estimators. Sensitivity analyses support assumptions underlying MR. The MR estimates largely aligned with estimates based on ordinary least squares of 5.9% (N=1,260,514) for the full population, sibling (5.3%, 95% CI=5.2-5.4; N=970,440) and monozygotic twin (3.2%, 95% CI=1.9-4.6; N=5,852) models. All estimates exceeded opportunity costs as proxied by the market interest rate, and thus suggests that education pays off, yet only MR circumvents the threat of unobserved confounding. Our study shows that the economic returns to schooling are substantial and demonstrates how genetic data can be used to investigate hard-to-answer causal questions.