The heterogeneous treatment effects of teenage childbearing on college attendance: In the case of the violation of the ignorability assumption


Park, Kiwoong (2016). The heterogeneous treatment effects of teenage childbearing on college attendance: In the case of the violation of the ignorability assumption. 2016 Add Health Users Conference. Bethesda, MD.


After accounting for confounding factors that affect both teenage childbearing and mother’s outcomes, the causal effect of teenage childbearing is unclear. Specifically, given that the effects of teenage childbearing may vary by women in different subgroups, it is necessary to measure the heterogeneous treatment effect to reveal the causal effects. One recent article, “The Effect(s) of Teen Pregnancy: Reconciling Theory, Methods, and Findings” in Demography journal (Diaz and Fiel 2016), used propensity score matching (smoothing-differencing method) to address the heterogeneity treatment effects and found that the negative effects of childbearing on college completion and early earning is most pronounced among women who are least likely to experience a teenage pregnancy. They also highlighted that the negative effects are significantly weaker among women most likely to become a teenage mother. Using Add Health data, this study replicates the findings of Diaz and Fiel (DF) and checks if the heterogeneous effects on education attainment (college attendance) is reliable. Since propensity score-based methods depends on the ignorability assumption, which means that treatment would be independent of potential outcomes by accounting for relevant covariates, it is crucial to include observable data to address selection into early fertility. However, DF’s analysis did not include extensive social support measurements, which may influence the probability of selection into teenage childbearing. Therefore, this study first includes common observable variables, following DFs study and previous study using Add Health (Kane et al. 2013), and then includes social support and mistreated experience in respondent’s childhood as covariates. In conclusion, this study partially speaks to the DF’s first finding that the negative effects of teenage childbearing on college attendance are most pronounced among women least likely to become teenage mothers. However, the result using new covariates shows strong negative effect of teenage childbearing for women most likely to experience an early fertility, refuting DF’s second finding that there is a weak but positive effect for most likely teenage mothers. These results suggest that the ignorability assumption of propensity score method is not likely to be verifiable, as Zhou and Xie discussed, and it is necessary to consider other methods together, like marginal treatment effects, to estimate the causal effects.


Reference Type

Conference proceeding

Book Title

2016 Add Health Users Conference


Park, Kiwoong

Year Published


City of Publication

Bethesda, MD

Reference ID