ABSTRACT
This chapter looks back to the foundation of randomization and ties together modern-day randomization approaches and considerations. Randomization removes, on average, the effect of any bias upon the experiment’s conclusions. Yet, in a single randomization, nontrivial imbalances in factors related to the outcome (i.e. confounders) can be common, and rare randomizations can fail dramatically. Restrictions to randomization help diminish the risk of rare, bad randomizations and can increase the efficiency of the experiment to detect the effect of an experimental condition, if one truly exists. The emphasis of this chapter is to understand the context of randomization, bring awareness of pros and cautions of various randomization schemes, and encourage covariate-adjusted randomization in experiments where experimental units are randomized sequentially. There is a wealth of additional literature on randomized trials where experimental units are all known before randomization and on covariate-adjusted response-adaptive randomization, though these topics are beyond the scope of this chapter.
