When Age Thresholds Fail: A Cautionary Tale About RDD Validity for Life-Course Outcomes

Abstract

Age-based policy thresholds have become popular instruments for regression discontinuity designs, but their validity depends on whether confounding characteristics are smooth at the cutoff. This paper demonstrates that age 26—a valid RDD threshold for studying insurance coverage—fails validity tests for life-course outcomes like fertility. Using 1.5 million observations from the American Community Survey (2011-2019), I document that private insurance coverage drops 4.0 percentage points at age 26 as young adults lose eligibility for parental coverage under the ACA. However, marriage rates—a key determinant of fertility—show a 5.6 percentage point discontinuity at the same threshold, violating the RDD identifying assumption. When stratified by marital status, neither married nor unmarried women show fertility discontinuities; the apparent pooled effect is entirely compositional. This methodological finding has broad implications: age thresholds that create sharp treatment variation often coincide with life-course transitions, making them unsuitable for studying outcomes like fertility, marriage, or labor supply that are mechanically correlated with age-related milestones. Applied researchers should treat balance tests as hard constraints, not supplementary diagnostics.