A comparison of conditional and unconditional randomization tests for highly stratified designs.

An unconditional randomization procedure is proposed for use in highly stratified randomized experiments where the sample size assigned to each treatment cannot be fixed in advance. This procedure is contrasted to the usual conditional randomization procedure in which sample sizes assigned within randomization units are treated as fixed and randomization units without both treatments assigned are deleted. The unconditional procedure is shown to have superior power for the populations studied.

RCT Entities:   Population Interventions Outcomes