How does early educational quality affect longer-term academic outcomes? We shed light on this question via a natural experiment in the Philippines--the implementation of a mother tongue education policy in public schools in kindergarten to Grade 3. This policy led to an unexpected decline in educational quality, but differentially in a subset of schools strongly predicted by pre-policy student language composition. We use language composition variables as instrumental variables for treatment. Leveraging panel data and confirming robustness to pre-trends, we find that the policy led to declines in standardized test scores in public primary schools. Employing a triple-difference strategy with Philippine Census data (across cohorts, localities, and decadal censuses), we show that by 2020, cohorts fully exposed to the policy completed 0.3 fewer years of schooling. By revealing how a policy-induced reduction in early education quality reduces educational attainment in later years, our results underscore the importance of investing in the quality of education in the first years of schooling.