Assumptions sufficient for the identification of local average treatment effects (LATEs) generate necessary conditions which allow to refute instrument validity. The degree of violations of instrument validity likely varies across subpopulations. In this project, we use causal trees to search and test for local violations of the LATE assumptions in a data-driven way. While existing instrument validity tests are unable to detect local violations, our procedure does -- as we also demonstrate in our simulations. We apply the proposed test in two different settings, namely parental preferences for mixed sex composition of children and the Vietnam draft lottery.
01.08.2019 - 31.12.2020 / Health Econometrics
Instrument Validity Tests with Causal Trees