Estimating returns to education using the genetic lottery

Does more schooling cause higher lifetime earnings? Social scientists have long sought to determine the economic returns to schooling given its importance to individual life chances and public policy. Prior estimates are limited by unobserved confounding in observational studies and the focus of popular quasi-experimental studies on increases in schooling at only one particular age. Genotyped data offer an opportunity to address unobserved confounding and to estimate the returns to an additional year of schooling at any age by using quasi-randomly assigned genetic variants related to education as instrumental variables [Mendelian randomization (MR)]. We analyze comprehensive Norwegian population registries with career-long earnings data and genotyped data from the Norwegian Mother, Father and Child Cohort Study (MoBa). We employ three identification strategies for triangulation: ordinary least squares (OLS) with covariate-adjustment, sibling and twin fixed-effects models, and MR. Estimated returns to schooling are 8.0% in MR (N = 109,800) and 6.3% in sibling-MR (N = 18,666). Extensive sensitivity analyses suggest that MR results are robust even to large potential violations of MR assumptions, including pleiotropy. MR estimates somewhat higher returns than OLS for the full population (5.9%, N = 1,255,604) and fixed-effects models for siblings (5.3%, N = 966,976) or monozygotic twins (3.2%, N = 2,630). The estimated internal rate of return to schooling exceeds opportunity costs of education as proxied by the market interest rate. The lifetime returns to schooling are positive and substantial across all models.