Working Paper: NBER ID: w30291
Authors: Dan A. Black; Joonhwi Joo; Robert Lalonde; Jeffrey A. Smith; Evan J. Taylor
Abstract: We provide simple tests for selection on unobserved variables in the Vytlacil-Imbens-Angrist framework for Local Average Treatment Effects (LATEs). Our setup allows researchers not only to test for selection on either or both of the treated and untreated outcomes, but also to assess the magnitude of the selection effect. We show that it applies to the standard binary instrument case, as well as to experiments with imperfect compliance and fuzzy regression discontinuity designs, and we link it to broader discussions regarding instrumental variables. We illustrate the substantive value added by our framework with three empirical applications drawn from the literature.
Keywords: instrumental variables; selection bias; local average treatment effects
JEL Codes: C26; C52; C93
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| use of tests (C52) | identification of selection effects (C52) |
| LATE framework (C22) | effectiveness of IV methods (C36) |
| selection on unobserved variables (C34) | limitations of LATE framework (C22) |
| treatment (M53) | impact on compliers (C59) |
| selection effects (C52) | generalizability of treatment effect estimates (C90) |
| conditional mean independence tests (C12) | insights into selection (D79) |
| selection on economic behavior (D01) | valid inferences about treatment effects (C90) |