A0825
Title: Joint diagnostic test of regression discontinuity designs: multiple testing problem
Authors: Koki Fusejima - The University of Tokyo (Japan) [presenting]
Takuya Ishihara - Tohoku University (Japan)
Masayuki Sawada - Hitotsubashi University (Japan)
Abstract: Diagnostic tests with a large number of covariates have been a norm to validate regression discontinuity (RD) designs. Such a procedure lacks its validity because of the multiple testing problem. Testable restrictions are to verify a single identification restriction, and therefore a single joint null hypothesis should be tested. In a meta-analysis of economics top five publications, the joint null was over-rejected and the null distribution of test statistics is distorted possibly by publication bias. We provide joint testing procedures based on the newly shown joint asymptotic normality of RD estimators. Simulation evidence demonstrates their favorable performances over Bonferroni correction for dimensions fewer than 10 covariates. However, neither Bonferroni correction nor our procedure guaranteed its size control with a larger number of covariates.
