Randomization

Randomization is the process of assigning study participants to different groups (such as treatment and control) using a chance mechanism, ensuring that each participant has a known, typically equal, probability of being placed in any group. It is the defining feature of the randomized controlled trial and, together with allocation concealment, the primary method for preventing selection bias in group assignment and for balancing both known and unknown confounders across groups at baseline on average, although chance imbalances can remain, especially in small trials.

Randomization is considered the gold standard for establishing causal relationships between interventions and outcomes. By assigning participants to groups based on chance rather than judgment or participant preference, randomization ensures that pre existing differences between the groups arise only by chance, so differences in outcomes larger than chance would explain can be attributed to the intervention rather than to those pre existing differences. Unlike statistical adjustment methods, which can only control for measured confounders, randomization balances both measured and unmeasured confounders in expectation. This property is why well conducted randomized controlled trials rank above other individual study designs in most evidence hierarchies for treatment effects (systematic reviews of such trials usually rank higher still) and why they are central to evidence based medicine.

Simple randomization assigns participants to groups with equal probability at each allocation, typically using a computer generated random number sequence. While it produces balanced groups in large samples, it can result in unequal group sizes in small studies.

Block randomization ensures that group sizes are balanced after every set of a specified number of allocations (the block size). For example, with a block size of four in a two arm trial, every four consecutive participants will include two in each arm. Variable block sizes are recommended to make the next assignment harder to predict, because with a fixed block size that becomes known, the last assignments in each block can be deduced.

Stratified randomization divides participants into strata based on important prognostic variables (such as age group, disease severity, or study site) and performs block randomization within each stratum. This improves balance on the stratification variables, even in smaller studies, provided the number of strata stays small relative to the number of participants; with many strata, many blocks are left incomplete and the balance is lost.

Minimization is an adaptive method that assigns each new participant to the group that would minimize the overall imbalance across several prognostic factors simultaneously, often with a random element added so that assignments are harder to predict. While technically not pure randomization, it is widely accepted and can achieve better balance than stratified randomization when multiple factors must be controlled.

The randomization sequence must be generated independently of the enrollment process, and allocation concealment (preventing those enrolling participants from knowing the upcoming assignment) is essential to prevent selection bias. Even a properly generated random sequence can be subverted if the enrollment team can predict or access upcoming assignments.

Use randomization whenever it is ethically and practically feasible to assign participants to groups in an intervention study. It is required for randomized controlled trials and is the strongest method for achieving causal inference. Randomization is not appropriate when ethical concerns prohibit withholding treatment (such as denying a life saving intervention), when the intervention has already been applied to a population (as in a natural experiment), or when the study is purely observational. In these cases, quasi experimental designs with appropriate analytical methods provide alternatives for causal inference, though with weaker guarantees.

A common error is using improper methods of randomization, such as assigning by date of birth, hospital record number, or day of the week, which are predictable and do not produce truly random allocation. Another mistake is confusing randomization with random sampling; randomization refers to group allocation within a study, while random sampling refers to the selection of participants from a population. Failure to conceal the allocation sequence from enrolling staff is a critical flaw that can lead to conscious or unconscious selection bias, even with a properly generated sequence. Finally, reporting that participants were "randomized" without describing the method of sequence generation, allocation concealment, or implementation provides insufficient information for readers to assess the quality of the randomization.

A multicenter trial of a new analgesic uses stratified block randomization with variable block sizes of four and six, stratified by study site and baseline pain severity (moderate versus severe). A central web based randomization system generates the allocation sequence and conceals the assignment until after the participant is enrolled and baseline data are recorded. Of 300 participants, 152 are assigned to the treatment group and 148 to the control group. Baseline characteristics (age, sex, pain severity, comorbidities) are well balanced between groups, with all standardized mean differences below 0.10.