randomization

Randomization is the process of assigning units to treatments or groups by chance rather than by choice. It aims to produce comparable groups and enable valid statistical inference. Randomization reduces selection bias and helps balance known and unknown confounders, supporting causal interpretation in experimental studies. It is distinct from random sampling, which selects individuals from a population for observation.

Common schemes include simple randomization (each unit has equal probability of assignment); stratified randomization (randomization within subgroups defined by key covariates); block randomization (ensuring balance within blocks); and cluster randomization (randomizing groups or sites rather than individuals). Allocation concealment and blinding are practices to prevent foreknowledge of assignments and reduce bias during enrollment and treatment.

Applications span clinical trials, agriculture, psychology, and other social sciences. Randomization underpins many statistical analyses and inference methods, such as intention-to-treat analyses and permutation tests. Design considerations include ethical equipoise, sample size, and potential noncompliance, which can affect balance and interpretation.

Limitations include the possibility of imbalance in small samples and practical constraints that prevent perfect randomization. Randomization does not remove all sources of bias or confounding, but it provides a principled basis for causal inference when used with appropriate design, conduct, and analysis.