Balancing tests in stratified randomized controlled trials: A cautionary note

Full text
Author(s):	Firpo, Sergio ^[1] ; Foguel, Miguel N. ^[2] ; Jales, Hugo ^[3] Total Authors: 3
Affiliation:	^[1] INSPER, Sao Paulo - Brazil ^[2] IPEA, Brasilia, DF - Brazil ^[3] Syracuse Univ, Syracuse, NY 13244 - USA Total Affiliations: 3
Document type:	Journal article
Source:	ECONOMICS LETTERS; v. 186, JAN 2020.
Web of Science Citations:	0
Abstract
We study the properties of imbalances tests of predetermined characteristics (covariates) in stratified randomized controlled trials. When the parameter of interest is a weighted average of the causal effects at all strata, then a test statistic based on a fixed effects regression model may suffice to detect the relevant imbalances. This will be true under a condition on the partial correlation between covariates and the outcome variable (uniformity condition). If that condition is violated, then the test based on the fixed effects regression will lack power. In that case, a test based on the fully saturated regression model can overcome this problem. Finally, we provide an omnibus test for the relevance of controlling for the covariates when estimating the average treatment effect (ATE). (C) 2019 Elsevier B.V. All rights reserved. (AU)

FAPESP's process:	11/02592-7 - An empirical likelihood estimator for average treatment effects in stratified randomized designs
Grantee:	Sergio Pinheiro Firpo
Support Opportunities:	Scholarships abroad - Research