Given a binary treatment D, a response Y and covariates X, often X interacts with D in an unknown way to make the treatment effect heterogeneous. If D is exogenous, there are various semi-parametric approaches (matching, inverse probability weighting, etc.) to estimate a weighted average of the heterogeneous effects without specifying the Y model. If D is endogenous, however, then there is hardly any practical semi-parametric approach available. With an instrument δ for endogenous D available, this paper proposes a simple instrumental variable estimator (IVE) without specifying the Y model. The IVE is consistent for the 'Cov(δ,D|X)-weighted average' of the heterogeneous effects, which happens to solve the 'weak instrument problem' because observations with a small Cov(δ,D|X) receives an accordingly small weight. If desired, an weighted IVE removing Cov(δ,D|X) can be also obtained, which is consistent for the (usual unweighted) average of the heterogeneous effects. The IVE is easy to implement with hardly any decision needed by the user. Also it has an asymptotic variance estimator that works well in small samples, and can be extended for multiple treatments. A simulation study and an empirical analysis are provided.
About the Speaker:
Prof. Myoung-jae Lee received his Ph.D. in Economics from University of Wisconsin-Madison in 1989 and is currently a Professsor of Economics at Korea University. His primary research interests are Econometrics and Labor Economics with the specific areas including Semiparametrics, Nonparametrics, Panel Data, Limited Dependent Variables, Sample Selection, Treatment Effects, Duration Analysis, etc.