Treatment Effects Using Inverse Probability Weighting and Contaminated Treatment Data: An Application to the Evaluation of a Government Female Sterilization Campaign in Peru
51 Pages Posted: 22 Sep 2016 Last revised: 23 Sep 2016
Date Written: September 21, 2016
We evaluate the impact of a female sterilization campaign implemented by the Peruvian government in 1996 and 1997 that we estimate impacted nearly 70,000 women. We use an inverse probability weighting (IPW) estimator that accounts for contamination in the available data. The contamination arises because while we observe sterilization status, we do not know if a given sterilization occurred as part of the campaign or whether it was chosen without influence from the campaign. The distinction is important because women targeted by the campaign and women who opted for sterilization outside of the campaign likely differ in many aspects, and we suspect the impact of sterilization is different for each group. We show that it is not necessary to fully observe whether a sterilized woman underwent the procedure because of the campaign to estimate unbiased average treatment effect of the government campaign. It is sufficient to estimate ― based on auxiliary data ― the conditional probability that if a sterilization is observed, it occurred because of the campaign. Using the proposed IPW estimator, we find that women sterilized because of the campaign had on average fewer 0.95 children. We also find substantial and statistically significant improvements in the height for age ― a measure of health ― of girls whose mothers were sterilized because of the campaign, and small but positive and statistically significant effects on years of schooling for boys.
Keywords: Female Sterilization, Fertility, Family Planning, Contaminated Data Models, Inverse Probability Weighting, Causal Effects, Observational Data
JEL Classification: C21, J13
Suggested Citation: Suggested Citation