Criteria-Based Randomization: Explicit and Exact Control in Multi-Arm Trials

26 Pages Posted: 13 Mar 2019

Date Written: July 1, 2018

Abstract

This paper introduces criteria-based randomization (CBR), a simple framework for controlling randomized trials. CBR uses only intervention assignments that the analyst believes will generate data useful for obtaining defensible causal estimates. In the multi-arm setting, CBR can create many groups which have equivalent pre-intervention covariate distributions. In this paper we compare CBR to several other multi-arm methods in the 4-arm and 12-arm setting. We assess the relative performance of the methods based on covariance balance as well as runtime and the mean squared error of estimators of the treatment effect. Further, we also show how CBR can be used in novel ways - creating intentionally imbalanced-designs so as to better assess components of a hypothesis. The CBR framework develops from previous work on tightening of trials in Tukey (1993) and Moulton (2004). CBR is also closely related to rerandomization (Morgan, Rubin, et al., 2012) but departs from rerandomization in scientifically important ways.

Keywords: statistics, causal inference, design of experiments, randomization, permutation test

Suggested Citation

Baiocchi, Michael and Kizilcec, René, Criteria-Based Randomization: Explicit and Exact Control in Multi-Arm Trials (July 1, 2018). Available at SSRN: https://ssrn.com/abstract=3339748 or http://dx.doi.org/10.2139/ssrn.3339748

Michael Baiocchi

Stanford University ( email )

Stanford, CA 94305
United States

René Kizilcec (Contact Author)

Cornell University ( email )

208 Gates Hall
107 Hoy Rd
Ithaca, NY 14853
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
68
Abstract Views
554
Rank
603,475
PlumX Metrics