Urn Randomization


Equivalence of patient groups is a critical issue in matching research where multiple treatments are implemented (Stout et al., 1994). Urn randomization was created to handle such complex research designs and “is systematically biased in favor of balance.” This type of randomization is only appropriate for large samples. The groundwork for the urn randomization was laid by Efron (1971) in the “biased-coin” design, while Wei (1978) presented the statistical theory behind urn randomization. According to Stout et al. (1994), urn randomization can “be used with many covariates, both marginally and jointly, producing optimal multivariate equivalence of treatment groups for large sample sizes” (p. 72).

A major disadvantage to urn randomization in the past has been the cost of implementing such a randomization procedure. Project MATCH developed a computer program for randomization, a prototype of which is now available free of charge to the general public. The documentation that follows offers a complete description of the program, its offerings, and procedures for obtaining the program. For further information about the use of urn randomization and particularly the use in Project MATCH, please refer to Stout et al. (1994).

Stout, R.L., Wirtz, P.W., Carbonari, J.P. & Del Boca, F.K. (1994). Ensuring balanced distribution of prognostic factors in treatment outcome research. Journal of Studies on Alcohol, Supplement No. 12, 70-75.

The Urn Randomization Program

The Urn Randomization Program is available as a DOS-based, generic randomization program. The program allows researchers to randomize study subjects to two or three randomization groups while balancing on two to 20 variables. A prototype procedure was used in Project MATCH, a large multisite clinical trial, and is discussed in Stout, et al. (1994).

Caution must be taken when setting up and using the procedure for the first time. Assignments to different conditions should be checked periodically to verify that the program is providing equivalent groups on all balancing variables.

Obtaining the Program

If you would like to obtain a FREE copy of the Urn Randomization beta program, send an email to vendetti@uchc.edu or write to:

Janice Vendetti, M.P.H.
Department of Public Health Sciences
UConn School of Medicine
263 Farmington Avenue, MC 6325
Farmington, CT 06030-6325 USA

Installing and Using the Program

Once you have obtained a copy of the program, you should consult and review the urn distribution instructions to learn how to install and use the Urn Randomization Program.

Urn Distribution Instructions