Keywords: minimization, randomization, rerandomization, misrandomization
When a clinical trial randomization needs to be stratified on more than 2 to 3 factors, a minimization technique such as that given in Pocock and Simon (1975) may be employed when the overall sample size is small compared to the number of stratification levels. Proschan, et al (2011) indicate that in such cases, a rerandomization test needs to be used when the minimization is done with unequal allocation to control Type I error rate. This test is acknowledged to be inefficient (Proschan, et al 2011), but furthermore its efficiency and bias when misrandomizations occur has not been thoroughly explored. Though it has been shown (Proschan 2017) that misrandomizations do not affect the Type I error rate, the power loss has not been characterized. We use simulation to explore the effects of misrandomization on the power of the rerandomization test, including random misrandomizations and confounded misrandomizations, such as those when the same sites consistently make the same mistake.