Skip to main content

Random Assignment: Definition, Methods and Examples

Random assignment explained with a 10,000-study simulation: with 20 students, chance left a pretest gap above 3 points in 51% of studies; with 400, in 0.2%.

Random Assignment: Definition, Methods and Examples

Random assignment is the use of a chance procedure, such as a computer-generated list, to decide which participants in a study receive which condition. Chance, not anyone’s choice, puts each participant in a group, so the groups differ only by chance before the treatment starts. This guide covers the methods, a simulation of 10,000 studies, the 1954 polio vaccine trial and how to report the procedure.

Key Takeaways

  • Assigning at random decides who gets which condition. Random sampling decides who is in the study. A study can use both, one or neither.
  • Groups formed by chance differ only by accident, even on characteristics nobody measured.
  • In our simulation, chance left a pretest gap above 3 points (0.3 standard deviations) in 51.0% of studies with 20 students, 13.5% with 100 and 0.2% with 400.
  • When simulated students chose their own group, a true 5-point effect came out at 11.5 to 11.6 points on average, at every sample size.
  • Report who generated the sequence and how, the type of randomization, how it was concealed, and a baseline table.

What is random assignment?

Random assignment means using a chance process to decide which condition each participant in a study receives. In its strict form, every participant has an equal chance of each condition, and each person is assigned independently of the others. Neither the researcher nor the participant has any say in who goes where. It is the step that turns a comparison of groups into an experiment.

The open textbook Research Methods in Psychology defines it as “using a random process to decide which participants are tested in which conditions”. Medicine calls the same step randomization or random allocation. The group that receives no treatment, or the standard one, is the control group.

Altman and Bland, in a BMJ statistics note, write that random “does not mean the same as haphazard”: each person has a known chance of each treatment, “but the treatment to be given cannot be predicted”. Alternating between groups, or assigning by date of birth or record number, fails that test, because whoever enrolls the next person already knows which group that person would join.

Random assignment vs random sampling

Random sampling decides who gets into a study; random assignment decides what happens to them once they are in. Sampling concerns whether the participants represent a population, so it affects how far the results generalize. Assignment concerns whether the groups are comparable, so it affects whether the treatment caused the difference between them. Random selection is another name for random sampling.

Random sampling (random selection)Random assignment
Question it answersWho is in the study?Who gets which condition?
What it supportsGeneralizing from the sample to the populationConcluding that the treatment caused the difference
Without itThe sample may not resemble the populationThe groups may differ in systematic ways before the treatment starts

OpenStax’s Psychology 2e defines a random sample as “a subset of a larger population in which every member of the population has an equal chance of being selected”; the glossary has an entry on random sampling. A study can have both, one or neither; the four examples below were written for this guide.

Assigned at randomNot assigned at random
Random sampleA random sample of voters, each shown one of two question wordings by chance: a causal claim about the populationA poll of a random sample that compares car owners with non-owners: describes the population, no causal claim
Not a random sampleA lab experiment or clinical trial with volunteers: a causal claim about people like the participantsTwo existing classes, one given a new textbook: neither claim is safe

Most psychology experiments are the volunteer kind, assigned at random but not randomly sampled: Research Methods in Psychology says that random sampling “is rarely used in psychological research”.

Why is random assignment important?

It takes the decision about who gets the treatment out of human hands. When researchers or participants choose, the groups differ in ways tied to that choice, such as motivation, health or prior ability, and those differences get mixed into the result. Chance has no preferences, so any difference between the groups at the start is an accident, and statistics can say how large an accident to expect.

The explanation paper for the CONSORT 2025 reporting guideline lists “three major advantages”. Randomization removes selection bias, “balancing both known and unknown prognostic factors”. It lets probability theory say how likely a difference in outcome is to reflect chance. And it makes blinding easier. The authors rank the first as usually the most important.

Altman and Bland describe the alternative: if the clinician chooses, “there will probably be differences in the clinical and demographic characteristics of the patients receiving the different treatments. Much the same will happen if patients choose their own treatment”. A characteristic that differs between the groups and also affects the outcome is a confounder, and chance is the only control that reaches the ones nobody measured. Our guide to confounding variables covers the other controls, and the one on correlation vs causation explains why a randomized experiment is the strongest single design for a causal claim.

How do you randomly assign participants?

Generate the assignment sequence by computer before recruitment starts, keep it hidden from whoever enrolls participants, and give each new participant the next assignment in the sequence. Simple randomization suits large studies. Small studies do better with block randomization, which keeps the group sizes equal, or stratified randomization, which also balances a characteristic you name in advance.

MethodHow it worksUse it whenWatch for
Simple randomizationEach participant is assigned independently, like a coin tossThe sample is largeUnequal group sizes and chance imbalance in small samples
Block randomizationWithin each short block, every condition appears equally often, in random orderGroup sizes must stay equalA known block size makes the last assignments in a block predictable
Stratified randomizationA separate blocked sequence for each subgroup (stratum), such as each siteA characteristic strongly predicts the outcome and the sample is smallThe number of strata multiplies quickly
Cluster randomizationWhole groups, such as classes or clinics, are assignedThe intervention is a policy or service delivered to whole groupsMembers of a cluster resemble each other, and the analysis must account for it

A second note by Altman and Bland, “How to randomise”, gives the working rules. With blocks, “The numbers in the two groups at any time can never differ by more than half the block length.” Stratified lists must be blocked within each stratum, and in small studies “it is not practical to stratify on more than one or perhaps two variables”. Stratified randomization is not stratified sampling, which uses strata to choose who enters the sample.

The CONSORT 2025 explanation paper trusts simple randomization in trials “of several hundred participants or more” and suggests a restricted method, such as blocks or strata, for those “of fewer than around 200 participants”. Blocking has a price: “Improved balance comes at the cost of reducing the unpredictability of the sequence.” For the last row of the table, Hemming and colleagues cover the design of cluster trials.

Allocation concealment means the person enrolling a participant cannot know the next assignment; the CONSORT authors call it “a critical mechanism that prevents foreknowledge of treatment assignment”. Trials use a central computer or “sequentially numbered, opaque, sealed envelopes”. In a student experiment, let the survey software reveal the condition only after consent.

The whole procedure in four steps:

  1. Fix the conditions and the allocation ratio (usually 1:1), and choose simple, block or stratified randomization.
  2. Generate the whole sequence by computer before recruiting, and record the seed.
  3. Keep the sequence away from anyone who enrolls participants.
  4. Give each participant the next entry, record it, and never reassign.

A worked example: how well does chance balance two groups?

We simulated a study-skills course with a true effect of 5 points on a final exam and formed the two groups in three ways. When chance alone formed the groups, the pretest gap between them averaged 3.6 points in one direction or the other with 20 students, 1.6 with 100 and 0.8 with 400, and the estimated effect averaged 5.0 points at every size. Under self-selection the estimated effect averaged 11.5 to 11.6 points.

These are simulated data, not real students. Each student has a pretest score (mean 60, standard deviation 10). The final exam score is 20 + 0.8 × pretest, plus 5 points for students in the course, plus noise with a standard deviation of 6, so final scores within a randomly formed group have a standard deviation of 10. We ran 10,000 studies at each size in Python (standard library, seed 20261121) and formed the groups by:

  • a random split into equal groups: shuffle the list and put the first half in the course (in the terms of the methods table above, block randomization with one block the size of the whole study);
  • matched pairs: rank students by pretest, pair neighbors, and flip a coin within each pair;
  • self-selection: each student opts in with a probability that rises with the pretest score: .27 at a score of 50, .50 at 60 and .73 at 70.
Students per study, assigned at randomAverage size of the pretest gap, in either directionStudies with a gap above 3 pointsAverage estimated effect (points)Middle 95% of estimates (points)
20 (10 per group)3.6 points51.0%5.0−3.8 to 13.6
100 (50 per group)1.6 points13.5%5.01.1 to 9.0
400 (200 per group)0.8 points0.2%5.03.0 to 7.0

Randomization was unbiased at every size: the estimates centered on the true 5 points. With 20 students, half of all studies began with groups more than 3 points apart on the pretest, and 13.4% ended with an estimated effect of zero or less.

Students per study, self-selectionAverage pretest gap (course minus comparison)Average estimated effect (points)Middle 95% of estimates (points)Studies with an estimate above 10 points
20+8.2 points11.53.2 to 20.063.5%
100+8.3 points11.67.9 to 15.380.2%
400+8.3 points11.69.8 to 13.595.8%

Under self-selection the stronger students chose the course, the pretest gap stayed at about 8 points, and the course looked more than twice as effective as it was. More data made one thing worse: the estimates gathered more tightly around the wrong answer. With 400 students, 95.8% of studies put the effect above 10 points.

Matched pairs, which amount to strata of two students, show why small studies stratify. With 20 students the average size of the pretest gap fell from 3.6 points to 0.9, only 0.6% of studies had a gap above 3 points, and the middle 95% of estimates narrowed to −0.4 to 10.7.

A real example: the 1954 Salk polio vaccine trial

The 1954 field trial of Jonas Salk’s polio vaccine ran two designs side by side. In 84 test areas in 11 states, children whose parents volunteered them were assigned at random to vaccine or placebo. In 127 test areas in 33 states, second graders whose parents agreed received the vaccine and were compared with all first and third graders. The randomized comparison showed the larger benefit.

The design counts are from Marcia Meldrum’s BMJ history of the trial. In Paul Meier’s account, the randomized part used “a carefully applied chance process that gave each volunteer an equal probability of getting vaccine or salt solution”, and each vial carried only a code number until every case had been diagnosed. The rows below come from Meier’s Table 1, adapted from the 1955 evaluation report by Thomas Francis Jr. and colleagues.

DesignGroupChildrenPolio casesCases per 100,000
Randomized, placebo-controlledVaccinated200,7455728
Placebo201,22914271
Not inoculated338,77815746
Observed controlVaccinated second graders221,9985625
Controls: all first and third graders725,17339154
Second graders not inoculated123,6055444

The not-inoculated row for the randomized areas includes 8,577 children who had one or two placebo injections; Meier’s rows for incomplete vaccinations are left out.

In the randomized areas the polio rate of the vaccinated children was about 60% lower than that of the placebo children. In the observed-control areas the difference was about 53%. The placebo row points to the reason. Children whose parents volunteered and who received only salt solution had 71 cases per 100,000; children in the same areas who were not inoculated had 46.

Meier calls this the volunteer effect and says it is “clearly evident” in the table. Families who volunteered, he reports, “tend to be better educated and, generally, more well-to-do”. Their children also had the higher polio rate: the 71 against 46 in the table. The observed-control design compared vaccinated volunteers with all first and third graders, whether their parents would have volunteered or not, so the benefit it showed was smaller.

In our simulation self-selection inflated the effect; in the polio trial it shrank the measured benefit, and nobody could have said which in advance. Meier’s verdict on the non-randomized arm: “Were the observed control information alone available, considerable doubt would have remained about the proper interpretation of the results.”

What does randomization not do?

It does not guarantee equal groups in any single study, it does not protect the comparison once people drop out or switch groups, and it says nothing about whether the results apply beyond the participants. Randomization removes bias from the way groups are formed; what happens after that is outside its reach.

Chance imbalance. The CONSORT 2025 explanation paper is direct: proper randomization “does not guarantee similarity of the groups at baseline. Any differences in baseline characteristics are, however, the result of chance rather than bias.” The safeguards are a restricted method in small samples and a baseline table. A significance test of the baseline gap is not one of them: “Significance testing of baseline differences is not recommended and should not be reported.” Such a test asks whether the gap arose by chance, and under proper randomization it did.

Attrition. The protection covers the groups as assigned. If a third of the students assigned to a course drop out, the ones who remain have selected themselves. The CONSORT rule: “To preserve the benefits of randomisation, all randomised participants are included in the analysis and retained in the group to which they were allocated.” This is an intention-to-treat analysis.

Generalizing. Participants in a psychology experiment are rarely a random sample, and the polio trial shows how much volunteers differ from everyone else. Extending a result to a wider population takes random sampling or a separate argument.

What if you cannot randomly assign?

Then you compare groups that formed some other way, and you say so. Some variables cannot be assigned at all, such as a person’s sex, and others cannot be assigned ethically, such as exposure to harm. The study becomes a quasi-experiment or an observational study, any causal claim is weaker or is replaced by a claim of association, and the work shifts to measuring and adjusting for the ways the groups differ.

OpenStax’s example of the ethical limit is a study of whether childhood abuse lowers adult self-esteem: it would require assigning some participants to be abused, and “that experiment would be unethical.” Altman and Bland’s advice for these cases is “to retain all the methodological features of a well conducted randomised trial other than the randomisation.”

Our guide to types of research methodology sets out which claims quasi-experimental and observational designs support, and the glossary has an entry on experimental design.

How do you report randomization in a paper?

Say who generated the allocation sequence and with what tool, what type of randomization you used and any blocking or stratification, and how the sequence was kept from the people enrolling participants. Then show a baseline table. A reader should be able to tell that nobody could have steered anyone into a group.

For health research the standard is the CONSORT 2025 statement, a 30-item checklist that replaces CONSORT 2010. Its randomization items, and the item for the baseline table, are these.

ItemTopicWhat to report
17aSequence generation“Who generated the random allocation sequence and the method used”
17bSequence generation“Type of randomisation and details of any restriction (eg, stratification, blocking and block size)”
18Allocation concealment mechanism“Mechanism used to implement the random allocation sequence (eg, central computer/telephone; sequentially numbered, opaque, sealed containers), describing any steps to conceal the sequence until interventions were assigned”
19Implementation“Whether the personnel who enrolled and those who assigned participants to the interventions had access to the random allocation sequence”
25Baseline data“A table showing baseline demographic and clinical characteristics for each group”

CONSORT is written for clinical trials, but the same facts serve an experiment in psychology or education.

The passage below was written for this guide as an illustration and describes an invented study.

Methods (illustration). Participants were randomly assigned in a 1:1 ratio to the study-skills course or the comparison condition. A researcher with no role in recruitment generated the allocation sequence by computer, using permuted blocks of four and six in random order, stratified by pretest score (below 60, or 60 and above). The sequence was stored in the survey platform, where the two assistants who enrolled students could not see it, and each student’s condition was revealed only after consent and the pretest were complete. All 100 randomized students were analyzed in the group to which they had been assigned.

Sentence two answers items 17a and 17b, sentence three items 18 and 19, and the last says that nobody was dropped or moved after assignment. Our guide on how to write a methodology section includes a published Method section that names the exact randomizer setting used.

Where Papero fits

Papero is an AI writing workspace for students and researchers. It is a writing and citation tool, not a randomization tool or a statistics package: it does not assign participants or calculate your statistics. It helps with the sentences that report the procedure.

Paste a DOI to cite the reporting guideline you followed, use Find papers to search more than 250 million papers through OpenAlex for studies that used the design you plan, and let Review’s Claims check find sentences that still need a citation. The thesis writing page shows how chapter outlines, the source library and Review fit together.

Frequently Asked Questions

Why do we use random assignment?

Researchers assign participants to conditions at random so that nobody’s choice decides who gets the treatment. Any other way of forming groups lets motivation, health or ability differ between them before the study starts, and those differences cannot be separated from the effect of the treatment. Groups formed by chance differ only by accident, including on characteristics nobody measured, and statistics can say how large an accident to expect.

What is the difference between random sampling and random assignment?

Random sampling selects who takes part in a study: every member of a population has an equal chance of being selected, which supports generalizing the results to that population. Assigning at random happens afterwards: the people already in the study are placed into conditions by chance, which supports a cause-and-effect conclusion. A study can use both procedures, one or neither.

Is random assignment used in correlational studies?

No. A correlational study measures variables as they occur and assigns nobody to anything, so there are no conditions to randomize. Assigning participants by chance, to conditions or to the order of conditions, is what makes a study an experiment; the textbook Research Methods in Psychology says that a study without it “is not an experiment.”

What is an example of random assignment in psychology?

A researcher testing whether a study-skills course raises exam scores recruits 100 students and lets a computer place 50 in the course and 50 in a comparison condition. OpenStax’s Psychology 2e uses a similar case: a sample of about 200 algebra students, each assigned by software to learn from a computer program or from an in-person teacher.

Does randomization guarantee that the groups are equal?

No. Randomization removes bias from the way groups are formed, but in any single study chance can still leave the groups unequal. In a simulation of 10,000 two-group studies with 20 students each, the groups differed on a pretest by more than 0.3 standard deviations in 51% of studies; with 400 students each, they did so in 0.2%.

Sources

  • Altman, D. G., & Bland, J. M. (1999a). How to randomise. BMJ, 319(7211), 703–704. https://doi.org/10.1136/bmj.319.7211.703
  • Altman, D. G., & Bland, J. M. (1999b). Treatment allocation in controlled trials: Why randomise? BMJ, 318(7192), 1209. https://doi.org/10.1136/bmj.318.7192.1209
  • Hemming, K., Eldridge, S., Forbes, G., Weijer, C., & Taljaard, M. (2017). How to design efficient cluster randomised trials. BMJ, 358, Article j3064. https://doi.org/10.1136/bmj.j3064
  • Hopewell, S., Chan, A.-W., Collins, G. S., Hróbjartsson, A., Moher, D., Schulz, K. F., Tunn, R., Aggarwal, R., Berkwits, M., Berlin, J. A., Bhandari, N., Butcher, N. J., Campbell, M. K., Chidebe, R. C. W., Elbourne, D., Farmer, A., Fergusson, D. A., Golub, R. M., Goodman, S. N., . . . Boutron, I. (2025a). CONSORT 2025 explanation and elaboration: Updated guideline for reporting randomised trials. BMJ, 389, Article e081124. https://doi.org/10.1136/bmj-2024-081124
  • Hopewell, S., Chan, A.-W., Collins, G. S., Hróbjartsson, A., Moher, D., Schulz, K. F., Tunn, R., Aggarwal, R., Berkwits, M., Berlin, J. A., Bhandari, N., Butcher, N. J., Campbell, M. K., Chidebe, R. C. W., Elbourne, D., Farmer, A., Fergusson, D. A., Golub, R. M., Goodman, S. N., . . . Boutron, I. (2025b). CONSORT 2025 statement: Updated guideline for reporting randomised trials. BMJ, 389, Article e081123. https://doi.org/10.1136/bmj-2024-081123
  • Jhangiani, R. S., Chiang, I.-C. A., Cuttler, C., & Leighton, D. C. (2019). Experimental design. In Research methods in psychology (4th ed.). Kwantlen Polytechnic University (kpu.pressbooks.pub, Wayback Machine capture of January 22, 2026): https://kpu.pressbooks.pub/psychmethods4e/chapter/experimental-design/
  • Meier, P. (1972). The biggest public health experiment ever: The 1954 field trial of the Salk poliomyelitis vaccine. In J. M. Tanur, F. Mosteller, W. H. Kruskal, R. F. Link, R. S. Pieters, & G. R. Rising (Eds.), Statistics: A guide to the unknown (pp. 2–13). Holden-Day. Read in the course copy at Loyola University Chicago: https://webpages.math.luc.edu/~mgb/courses/s203/MeierPolio.htm
  • Meldrum, M. (1998). “A calculated risk”: The Salk polio vaccine field trials of 1954. BMJ, 317(7167), 1233–1236. https://doi.org/10.1136/bmj.317.7167.1233
  • Spielman, R. M., Jenkins, W. J., & Lovett, M. D. (2020). Psychology 2e, Section 2.3, Analyzing findings. OpenStax. https://openstax.org/books/psychology-2e/pages/2-3-analyzing-findings

Papero Team

Papero

The Papero team builds an AI writing assistant for students, scholars and researchers, and writes about academic writing from what the product does.

Follow along on Instagram at @paperoai for campaign breakdowns.

Your next paper, with real citations.

Describe your paper and get an outline in a minute. The Free plan needs no card.

Write your paper with the sources attached.