Nontestability of instrument validity under continuous treatments
Author:
- 15 December 2020
- journal article
- research article
- Published by Oxford University Press (OUP) in Biometrika
- Vol. 108 (4) , 989-995
Abstract
Summary: This note presents a proof of the conjecture in Pearl (1995) about testing the validity of an instrumental variable in hidden variable models. It implies that instrument validity cannot be tested in the case where the endogenous treatment is continuously distributed. This stands in contrast to the classical testability results for instrument validity when the treatment is discrete. However, imposing weak structural assumptions on the model, such as continuity between the observable variables, can re-establish theoretical testability in the continuous setting.Keywords
This publication has 13 references indexed in Scilit:
- Erratum: On falsification of the binary instrumental variable modelBiometrika, 2017
- Testing multivariate economic restrictions using quantiles: The example of Slutsky negative semidefinitenessJournal of Econometrics, 2016
- Causal Inference for Statistics, Social, and Biomedical Sciences2015
- A Test for Instrument ValidityEconometrica, 2015
- Measure Theory2007
- Speakable and Unspeakable in Quantum Mechanics2004
- Retrospectives: Who Invented Instrumental Variable Regression?Journal of Economic Perspectives, 2003
- Classical Descriptive Set Theory1995
- Realization of the Einstein-Podolsky-Rosen paradox for continuous variablesPhysical Review Letters, 1992
- Proposed Experiment to Test Local Hidden-Variable TheoriesPhysical Review Letters, 1969