Comparing Effects When Access to Alternatives Varies: Revisiting ‘Bringing Schools to Afghan Girls’
Updated: Nov 16, 2020
Experimenters introduce treatments into settings characterized by pre-existing alternatives to the treatment. Treatment effects depend on whether subjects take up the new treatment or stick with one of the alternatives. Such take-up behavior varies across contexts and therefore contributes to variation in effects over replications in different contexts. Using a principal stratification approach, we develop methods to decompose treatment effects on the basis of subjects’ take-up behavior. This allows us to distinguish differences in access to treatment alternatives from differences in treatment efficacy as sources of effect heterogeneity across experimental replications. We demonstrate these methods using two replications of a field experiment on schooling options in rural Afghanistan.