Randomisation uses a chance-based procedure to assign experimental units or determine an order. It helps prevent investigator choices and other systematic patterns from determining the comparison. To assess it in a paper, the useful question is what was actually randomised and how, rather than whether the word random appears somewhere in the methods.
Distinguish random allocation from casual selection
NC3Rs’ Experimental Design Assistant distinguishes a documented random allocation strategy from haphazard assignment based on human judgement. It also treats allocation concealment as a complementary protection against selection bias.NC3Rs EDA — Allocation (opens in a new tab)
A statement that samples were selected at random may concern sampling from a stock, not assignment to experimental conditions. Those are different stages. Identify which one the authors mean.
| Target | Example of the design question |
|---|---|
| Experimental unit | Which unit receives each condition? |
| Position | Where is each unit placed? |
| Processing order | Which unit is handled or measured next? |
| Treatment sequence | In what order are conditions encountered? |
Alternating assignments in a predictable A–B–A–B pattern is not the same as generating them by a random process. A neat-looking layout can remain vulnerable to predictable allocation or time-related patterns.
Look for conditions confounded with position or time
NC3Rs’ DRIVER guidance notes that plate position, handling and processing order can influence in vitro results. Randomisation can address allocation and order-related sources of bias, depending on what the design actually randomises.NC3Rs — DRIVER recommendations: risk of bias (opens in a new tab)
In an original hypothetical experiment, all condition-A units occupy one side of a plate and all condition-B units the other. If the two regions differ in an environmental influence, condition and position are inseparable in that comparison.
Randomly assigning sample identifiers before placing all A units on one side does not solve that layout problem. The randomised step must address the factor that could otherwise track the condition.
Distinguish chance balance from designed balance
Small randomly assigned groups can differ on a baseline characteristic by chance. Randomisation is a property of the assignment process, not a guarantee that every observed characteristic will be equal in the realised groups.
The EDA describes randomisation within blocks to account for relevant sources of variability and notes that the analysis should reflect the blocking factor.NC3Rs EDA — Allocation (opens in a new tab)
For an original example, if an experiment is run over two days, assigning both conditions within each day can make the condition comparison less entangled with day than assigning one condition per day. The actual randomisation and analysis still need to be specified.
Do not infer good allocation from similar baseline averages alone. Deliberately choosing units to make those averages look alike and using a documented random allocation are different procedures with different implications.
Check the description and later departures
A useful report identifies the experimental unit, sequence-generation method, restrictions or blocks, and implementation. It should also explain relevant departures rather than describing only the intended design.
If units are excluded after allocation, ask why and whether exclusion decisions depended on the condition or observed outcome. A sound initial randomisation does not make selective later removal harmless.
Keep blinding separate. The person assessing an outcome may know the allocation even when it was generated randomly, so randomisation alone does not address expectation effects during measurement.
Finally, assess the comparison in the model studied. Random assignment within one experimental system strengthens particular internal comparisons; it does not by itself establish that the same effect will occur in another system or population.
Sources and further detail
- NC3Rs EDA — Allocation (opens in a new tab)
Allocation, haphazard assignment, blocking and analysis guidance read. No allocation sequence or laboratory procedure is generated.
- NC3Rs — DRIVER recommendations: risk of bias (opens in a new tab)
Randomisation discussion for in vitro allocation, position and processing-order effects read. Plate and two-day examples are original hypothetical illustrations.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.