Description:

Filip Obradovic (Northwestern University): Long-Term Treatment Effect Identification via Data Combination

Abstract: "Identification of long-term average treatment effects (LTE) is an important research goal, which is often infeasible via experimentation. Recent literature proposes combining short-term experimental and long-term observational data to provide credible alternatives to observational studies. The paper makes two distinct contributions in this context. I first show that experimental data have an auxiliary role. They bring no identifying power without additional assumptions in the observational data. When such assumptions are imposed, experimental data serve to amplify their power. If the assumptions fail, adding experimental data may only lead to results that are farther but never closer to the truth. Motivated by this, I introduce two assumptions on treatment response that may be defensible based on economic theory or intuition. To utilize them, I develop a novel two-step identification approach that centers on bounding temporal link functions -- the relationship between short-term and mean long-term potential outcomes. This approach allows for imperfect experimental compliance and provides sharp bounds on the LTE for a general class of assumptions -- extending existing results under compliance issues. I illustrate the findings by studying the long-term effects of Head Start participation. To do so, I create a new combined dataset using the Head Start Impact Study and the National Longitudinal Survey of Youth."