Does education really not matter for health?

In a recent paper, Albarran et al. (2020) use a well-known quasi-randomization strategy to study the impact of education on health outcomes. Concretely, they use the fact that changes in compulsory education laws-usually extending the age of release from the school system-result in exogenous differences in educational attainment between people who may only be separated by a few months of age. They deploy an Instrumental Variable (IV) strategy to compare the outcomes of people forced to stay longer in school to those of individuals who were allowed to drop out earlier. Their empirical strategy does not find a statistically significant effect of education on health outcomes. The paper is timely, well written, and well executed, and deserves careful attention. However, we hope readers consider four caveats-some already raised by the authors-before concluding from the results that education does not matter for health. To help readers interpret this solid research piece, we call attention to: 1) previous research on the links between education and health; 2) the selection of countries in the paper in the presence of heterogenous treatment effects; 3) biases generated by misclassification and measurement errors; and 4) the interpretation of local average treatment effects (LATE) arising from the policy interventions used as instrumental variables.