Estimating Program Impacts for a Subgroup Defined by Post-Intervention Behavior: Why is it a Problem? What is the Solution?

This brief demonstrates why estimating program effectiveness for an endogenous subgroup leads to compositional differences between the treatment and control groups that may result in biased impact estimates. Specifically, it shows the source of bias in the estimated impact of a teenage pregnancy prevention program on the likelihood that youth use contraception, among a subgroup of youth who are sexually active post–random assignment. It also provides guidance for meeting similar research objectives by constructing outcomes so that they yield unbiased estimates of program impacts.