Randomization and Blocking

Randomization is the allocation of participants to study groups by a chance mechanism, and blocking is a refinement that keeps the group sizes balanced as allocation proceeds. Randomizing makes the comparison groups exchangeable in expectation, so that differences in outcome can be attributed to the intervention rather than to systematic differences between groups. Together with allocation concealment, these methods are what give a randomized trial its strength as a tool for causal inference.

Definition

Randomization assigns participants to comparison groups using a chance mechanism so that the groups are expected to be similar in all factors except the assigned intervention, and blocking constrains the random sequence within successive blocks so that group sizes stay approximately equal throughout enrollment.

Scope

The entry covers why randomization controls both measured and unmeasured confounding, how allocation sequences are generated, how blocking maintains balance over time, and the related but distinct safeguards of allocation concealment and blinding. It is framed as a methodological reference on experimental design and is not a guide to conducting clinical care.

Key concepts

• Random allocation and exchangeability
• Control of measured and unmeasured confounding
• Permuted block randomization
• Allocation concealment
• Blinding (masking)
• Stratified and block randomization combined
• Predictability and selection bias at allocation

Mechanisms

By assigning intervention through chance, randomization breaks any link between a participant's characteristics and the group they enter, so the groups are balanced in expectation on both recorded and unrecorded factors; this is what licenses attributing outcome differences to the intervention. Blocking divides the allocation sequence into blocks within which equal numbers go to each group, preventing the chance imbalances in group size that can arise in small or interim samples; permuted blocks vary block size to limit predictability. Allocation concealment protects the sequence from being foreseen at the moment of assignment, and blinding protects later stages from differential treatment and assessment; these are distinct safeguards that randomization alone does not provide.

Clinical relevance

Whether and how a trial randomized, concealed allocation, and blocked or stratified its assignment are central to judging how much confidence its results deserve, since failures at this stage are a common source of exaggerated treatment effects. This entry describes trial methodology for appraisal and research and is not a basis for individual treatment decisions.

Evidence & guidelines

The CONSORT 2010 statement asks trials to report the method of sequence generation, any stratification or blocking, and how allocation was concealed, reflecting evidence that inadequate or unconcealed randomization is associated with biased effect estimates. The Lancet methodology series distinguishes sequence generation, allocation concealment, and blinding as separate protections, a distinction that is widely used in risk-of-bias appraisal.

History

Randomization, replication, and blocking were introduced by R. A. Fisher in the 1920s and 1930s for agricultural experiments, where blocking controlled field heterogeneity. Austin Bradford Hill brought randomization into medicine in the streptomycin trial of the late 1940s, establishing the randomized controlled trial as a clinical standard. Later methodological work, consolidated in reporting guidelines such as CONSORT, clarified the separate roles of sequence generation, blocking, allocation concealment, and blinding.

Debates

Does block randomization create predictability?

Fixed, small block sizes can let an unblinded investigator guess upcoming allocations near the end of each block, reintroducing selection bias; varying block size (permuted blocks) and concealing allocation are used to counter this, trading some balance for unpredictability.

Is randomization enough on its own?

Randomization balances groups at baseline but does not by itself prevent foreknowledge of assignment or differential handling afterward; allocation concealment and blinding are separate safeguards, and a trial can be randomized yet still biased if these fail.

Key figures

• Ronald A. Fisher
• Austin Bradford Hill
• Kenneth Schulz
• David Grimes
• Douglas Altman

Seminal works

• fisher-1935-doe
• schulz-grimes-2002-generation
• schulz-grimes-2002-concealment

Frequently asked questions

Why does randomization control confounding when matching and stratification only control known factors?

Because assignment is by chance, randomization tends to balance the groups on all characteristics, including ones that were never measured or suspected, whereas matching and stratification can only balance the specific factors chosen in advance.

What is the difference between randomization, allocation concealment, and blinding?

Randomization is the chance mechanism that creates the groups; allocation concealment hides the upcoming assignment so it cannot be foreseen or manipulated at enrollment; blinding keeps participants, caregivers, or assessors unaware of group assignment after randomization. They are distinct safeguards and are reported separately.