Answers are grounded in this paper and cite their sources.
Under the exclusion restriction, independence, and monotonicity assumptions, Imbens and Angrist (1994) show that the Wald estimand is equal to the local average treatment effect (LATE). However, these assumptions entail additional restrictions, which we exploit for inference on the LATE. Specifically, we propose a conditional likelihood ratio test that we show controls asymptotic size uniformly over a large class of data-generating processes. The test has very low computational cost: it requires no simulation and involves solving at most four quadratic programming problems. Monte Carlo simulations show that our test can offer substantial power gains over the Wald and Anderson-Rubin tests, including when the instrument is weak.
No takes yet. Share an insight, caveat, or question.
D’Haultfœuille et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: