13 Research methods, statistics, critical review and evidence-based practice13.3 Experimental studies
13.3.1.4 Allocation concealment and randomisation methods
2 min read1 figure
Written by Resident Doctor Committee · Published
Randomisation and concealment
Random allocation acts at entry and is what makes the trial arms comparable at baseline.
Blinding, intention-to-treat analysis and multicentre recruitment act later or address other problems, so none of them creates baseline comparability.
Step | Protects against | When it acts | Possible in every trial? |
|---|---|---|---|
Sequence generation | A patterned or chosen order | Before anyone is enrolled | Yes |
Allocation concealment | Selection bias at entry | At enrolment, until allocation is revealed | Yes |
Blinding | Differences in care, reporting and outcome assessment | After allocation | No |
Allocation concealment means those enrolling participants cannot foresee the next random assignment.
Concealment works by holding the sequence away from recruiters and releasing each assignment, for example through a remote central service, only once a participant is enrolled.
Quasi-randomisation
Quasi-randomisation allocates by a fixed, predictable rule: alternation, date of birth, hospital number or day of the week.
It is not true randomisation, because recruiters can foresee the next assignment, so allocation cannot be concealed and selection bias can enter.
An openly accessible random list has a different fault: the sequence is random but concealment has failed.
Methods of random allocation
Method | How it allocates | What it achieves |
|---|---|---|
Simple | Each participant independently by chance | Can drift out of balance by chance in small trials |
Block (permuted block) | Within fixed-size blocks | Arm sizes differ by no more than half a block at any point in recruitment |
Stratified | Separate sequences within a few prognostic strata, often blocked | Balance on chosen factors is forced by design, not left to chance |
Minimisation | Each new participant goes to the arm that best keeps prognostic factors balanced against those already allocated | The next assignment depends on earlier participants' characteristics |
Cluster | Whole groups such as wards, teams or practices | Keeps the arms apart and limits contamination through shared contact |
Block randomisation
Under simple 1:1 randomisation of 40 people, a 25:15 split or worse occurs about 15% of the time.
Fixed block sizes are guessable: in a block of four, after A, B, B the last allocation must be A.
Randomly varying, undisclosed block sizes such as four, six and eight, with central allocation, keep assignments unpredictable.
Stratified randomisation
Randomisation with concealment still gives each participant an equal probability of each arm and protects against selection bias; stratification adds forced balance on the chosen factors.
Too many stratification factors leave sparse strata, so trials stratify on only a few.
Minimisation
Minimisation balances marginal totals rather than every combination of factors, so it handles more factors than stratification.
A random element, such as a 0.8 probability of the preferred arm, makes it less predictable.
Cluster randomisation
Similarity within clusters reduces the effective sample size, so trials plan for the design effect and use cluster-aware analysis.