Estimating the educational consequences of teenage childbearing: Identification, heterogeneous effects and the value of biological relationship information.

Understanding the contribution of childbearing to social disadvantages of teenage mothers requires estimates that control for unobservables and generalize to teenage mothers. Sibling-differences and Instrumental Variables (IV) are common approaches to this end. Using the "Add Health" data, which oversampled siblings, and building on IV specifications from a widely-cited study, we compare various estimates of the consequences of teenage childbearing for schooling attainment. These IV-based estimates suggest moderate to large adverse impacts of teenage births (point estimates of -0.7 years of schooling or larger). However, the IV estimates are highly sensitive to choice of instrument and model specification. Estimates based on sibling and twin differences are consistently near zero-e.g., an estimated difference of -0.1 years between a teen mother and her biological full sister who did not have a teen birth-and are estimated with sufficient precision to exclude effects larger than -0.5 years. We review concerns about sibling methods and conclude that, despite their limitations, sibling estimates should be admitted along with other evidence on the consequences of teenage childbearing. Appreciation of the sensitivity of IV estimates and their other limitations would reinforce this conclusion.