What is a controlled experiment?
By Mubarak MutesasiraUpdated 12 min read
Short answer:
A controlled experiment is one in which the researcher changes only one factor, the independent variable, keeps all other conditions the same, and compares the result with a control group or condition that did not get the change. Because the groups differ only in that one factor, a difference in the outcome can be attributed to it. Randomly assigning participants to groups and blinding them to their group make that conclusion stronger.
This guide covers the parts of a controlled experiment, the difference between a control group and controlled variables, randomization, blinding and placebos, six worked examples from school science to clinical trials, and what stops an experiment from being controlled. It is part of our types of research design series.
What are the parts of a controlled experiment?
A controlled experiment has a hypothesis, an independent variable that the researcher changes, a dependent variable that is measured, controlled variables held constant, and at least one comparison group.
| Part | What it is | Plant example (illustrative) |
|---|---|---|
| Hypothesis | A testable prediction | Bean plants given fertilizer grow taller than plants without |
| Independent variable | What the researcher changes | Fertilizer (none, 5 g, 10 g per week) |
| Dependent variable | What is measured | Plant height after 4 weeks (cm) |
| Controlled variables | Conditions kept the same for all groups | Soil, pot size, light, water, temperature, seed variety |
| Control group | Receives no treatment, or the standard one | Plants with no fertilizer |
| Experimental group(s) | Receive the treatment | Plants with 5 g or 10 g |
| Replication | Several units per group | 10 plants per group |
| Random assignment | Chance decides which unit goes where | Draw seed numbers to assign pots |
If you are unsure which variable is which, our guide to independent vs dependent variables has a quick test and a quiz.
Control group vs controlled variables
A control group is a comparison group that does not get the treatment; controlled variables are conditions held the same for every group. Students and examiners often confuse them, but they do different jobs.
| Control group | Controlled variables | |
|---|---|---|
| What it is | A set of participants or samples | A set of conditions |
| Job | Shows what happens without the treatment | Stops anything else from explaining the result |
| Without it | You cannot tell whether the treatment did anything | Another factor might explain the difference |
| Example (illustrative) | Pupils taught with the usual method | Same test, same time of day, same marking rubric |
There are several kinds of control group:
- No-treatment control: gets nothing. Common in school and lab science.
- Placebo control: gets an inactive version that looks the same as the treatment (see below).
- Active or standard-care control: gets the current best treatment. Used when withholding treatment would be unethical.
- Wait-list control: gets the treatment after the study ends. Common in education and psychology.
- Positive and negative controls: in lab work, a sample known to give a result and one known not to, to check the method works.
Why is randomization important?
Random assignment gives every participant the same chance of being in each group, so differences between people, measured or not, are balanced across groups on average. This is what lets a true experiment rule out the "the groups were different to begin with" explanation [1].
Common methods:
| Method | How it works | When to use it |
|---|---|---|
| Simple randomization | Each unit assigned by a random number or coin toss | Large samples |
| Block randomization | Assign in blocks (for example, of 4 or 6) so group sizes stay equal | Small samples, or recruitment over time |
| Stratified randomization | Randomize separately within strata, such as age group or site | When a known factor strongly affects the outcome |
| Cluster randomization | Assign whole groups (classes, clinics, villages) | When individuals in a group cannot be separated |
Two rules make randomization work:
- Use a real random process, such as a computer-generated list. Alternating, using birth dates or letting staff choose is not random [2].
- Conceal the allocation until each participant is enrolled, so the person recruiting cannot predict or change the next assignment. The CONSORT 2010 statement asks trials to report how the sequence was generated and how it was concealed [3].
Random assignment is not the same as random sampling. Random sampling decides who is in the study and affects how far you can generalize; see probability vs non-probability sampling. Random assignment decides which group each participant joins and affects whether you can claim cause.
What is blinding?
Blinding (or masking) means keeping people unaware of which group a participant is in, so expectations cannot change behavior, care or how outcomes are measured.
| Who is blinded | What it prevents |
|---|---|
| Participants | Changing behavior or reporting because they know their group |
| Those delivering the treatment | Treating groups differently |
| Outcome assessors | Scoring or recording results in favor of a group |
| Data analysts | Choosing analyses after seeing which group did better |
Terms like "single-blind" and "double-blind" are used inconsistently. When Devereaux and colleagues asked physicians what "double-blind" meant, they gave many different answers [4]. Schulz and Grimes recommend saying exactly who was blinded instead [5], and CONSORT asks for the same [3].
Blinding matters most for subjective outcomes, such as pain, mood or a teacher's rating. For objective outcomes, such as death or a machine reading, it matters less. When participants cannot be blinded (you cannot hide that someone is exercising), blind the people who measure the outcome.
What is a placebo?
A placebo is an inactive treatment made to look and feel like the real one, given to the control group so that both groups have the same experience apart from the active ingredient. It allows the study to separate the effect of the treatment from the effect of being treated and expecting to improve.
- In drug trials: a tablet with no active drug, of the same size, color and taste.
- In other fields: a "sham" procedure, such as a sham device that looks switched on, or an attention-matched session with no active content in education or psychology.
Placebo effects are real but uneven. A review of trials that compared placebo with no treatment found little evidence of large effects in general, with possible small effects on subjective outcomes such as pain [6].
Placebo controls have ethical limits. The Declaration of Helsinki says a new intervention should normally be tested against the best proven one, and allows a placebo mainly where no proven intervention exists or where there are compelling scientific reasons and patients are not put at risk of serious or irreversible harm [7]. Your ethics committee will judge this; see how to write an IRB application.
Controlled experiment examples
All examples below are illustrative unless a source is given.
1. School science: does salt change the boiling point of water?
- IV: salt added (0, 10, 20 g per liter). DV: boiling temperature (°C).
- Controlled: same pan, same volume of water, same heat setting, same thermometer, same starting temperature.
- Control: pure water (0 g). Replication: three trials per level, mean reported.
2. Psychology: does background music affect memory?
- 90 students are randomly assigned to learn a word list in silence, with instrumental music, or with lyrical music.
- IV: sound condition. DV: words recalled after 10 minutes.
- Controlled: same list, room, time and instructions. The researcher scoring recall does not know the condition (assessor blinded).
3. Agriculture: does a new seed variety yield more?
- A field is divided into 24 plots; 12 are randomly assigned the new variety and 12 the standard one.
- IV: variety. DV: yield (t/ha).
- Controlled: same planting date, fertilizer and irrigation. Randomizing plots matters because soil quality varies across the field.
4. Business: does a shorter checkout page raise sales?
- Website visitors are randomly shown the current five-step checkout or a new three-step one. This is an online controlled experiment, or A/B test [8].
- IV: checkout version. DV: share of visitors who complete a purchase.
- Controlled: same prices, products and time period, because both versions run at the same time.
5. Engineering: does a coating reduce corrosion?
- 40 steel samples are cut from one sheet; 20 are randomly chosen for the coating, 20 are left uncoated.
- IV: coating. DV: mass loss after 500 hours in a salt-spray chamber.
- Controlled: same steel batch, chamber, exposure time and cleaning method.
6. Clinical trial: does a drug lower blood pressure?
- Adults with high blood pressure are randomized by computer, with concealed allocation, to the drug or a matching placebo for 12 weeks.
- Participants, clinicians and outcome assessors do not know the allocation.
- IV: drug or placebo. DV: change in systolic blood pressure.
- Reported following CONSORT, with a flow diagram of everyone randomized [3]. Results of a trial like this inform care only after professional review of the full evidence.
A real historical example. In 1747, James Lind gave six pairs of sailors with scurvy six different treatments, with diet and quarters kept similar. The pair given oranges and lemons recovered fastest [9]. It was controlled, but not randomized or blinded, and with two sailors per group it would not meet a modern standard.
What makes an experiment not controlled?
An experiment is not controlled when something other than the independent variable differs between the groups, or when there is no fair comparison at all.
| Problem | Example (illustrative) | Fix |
|---|---|---|
| No comparison group | Test scores rise after a new app is introduced | Add a control group taught without the app |
| Before-and-after only | Pain falls after a treatment, but it may have fallen anyway | Compare with a group measured over the same period |
| More than one change at a time | New fertilizer and a sunnier windowsill | Change one variable; hold the rest constant |
| Participants choose their group | Volunteers for a fitness program are already fitter | Randomize |
| Groups treated differently | The treatment group gets extra staff attention | Match attention in the control group |
| Unblinded subjective outcome | The researcher who designed the intervention rates improvement | Blind the assessor |
| Too few units | One plant per group | Replicate; run a power calculation |
A study where the researcher cannot assign the treatment, such as comparing smokers and non-smokers, is observational, not experimental. One where the researcher assigns treatment but not at random is a quasi-experiment [1]. Both are useful, but neither can rule out confounding as well as a randomized controlled experiment. See levels of evidence for how these designs rank and what a cohort study is for the main observational design.
Checklist for designing a controlled experiment
- One clear, testable hypothesis
- The IV and its levels named; the DV measured with a defined, reliable instrument
- A list of variables to hold constant, and how
- A suitable control group (none, placebo, standard care or wait-list)
- Random assignment with a real random process, concealed until enrollment
- Blinding of everyone who can be blinded, stated by role
- Sample size from a power calculation
- A primary outcome and analysis plan written before data collection
- Ethics approval and, for trials, registration before the first participant
How to design a controlled experiment with AI for Research
- Frame the question. The PICO question builder turns population, intervention, comparison and outcome into a question and a search concept table. It is free, needs no account and runs in your browser.
- See how others did it (0 credits). In Find Papers, filter by study type, such as randomized controlled trial, to see which control groups, blinding and outcomes similar studies used.
- Draft the design (3 credits). The Methodology builder drafts design, sampling, measures, analysis and limits for your question, with citations to studies from its search.
- Analyze the results (free). The data analysis tool runs Welch and paired t-tests and chi-square tests on a CSV in your browser and writes each result in APA style.
The Free plan has 45 credits a month; see pricing. Limits: the tools do not randomize participants, check your blinding or approve your design. Clinical trial designs need review by your supervisor, a statistician and an ethics committee.
FAQ
A controlled experiment is one in which what?
A controlled experiment is one in which the researcher changes only the independent variable, keeps every other condition the same, and compares the outcome with a control group or control condition that did not receive the change. Because nothing else differs between the groups, a difference in the outcome can be attributed to the independent variable. Random assignment to groups makes that claim stronger.
What is the difference between a control group and a controlled variable?
A control group is a set of participants or samples that do not receive the treatment, so their outcome shows what happens without it. A controlled variable is a condition kept the same for every group, such as temperature, light or timing, so it cannot explain the result. A plant experiment has a no-fertilizer control group and controlled variables such as light, water and soil.
Does a controlled experiment need a control group?
Most do, because a comparison is needed to know what would have happened without the treatment. Some controlled experiments compare two or more active treatments instead of a no-treatment group, or test the same participants under every condition in a within-subjects design, where each person acts as their own control. What every controlled experiment needs is a fair comparison with all other conditions held constant.
What is the purpose of randomization in an experiment?
Random assignment gives each participant the same chance of being in each group, so differences between people, both measured and unmeasured, are spread evenly across groups on average. This means a difference in outcome is unlikely to be caused by who ended up in which group. It also stops the researcher from steering particular participants into the treatment group, consciously or not.
What is a double-blind experiment?
In a double-blind experiment, neither the participants nor the people giving the treatment or measuring the outcome know who is in which group. This prevents expectations from changing behavior or how results are recorded. Because 'double-blind' is used inconsistently, the CONSORT reporting guideline asks researchers to state exactly who was blinded: participants, care providers, outcome assessors or data analysts.
Next steps
- Name your variables: independent vs dependent variables.
- Compare the experiment with other designs: types of research design.
- Make sure your outcome measure works: reliability vs validity.
- Work out how many participants you need: statistical power and sample size.
Sources
- Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin.
- Schulz, K. F., & Grimes, D. A. (2002). Generation of allocation sequences in randomised trials: chance, not choice. The Lancet, 359(9305), 515–519. https://doi.org/10.1016/S0140-6736(02)07683-3
- Schulz, K. F., Altman, D. G., & Moher, D., for the CONSORT Group. (2010). CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials. BMJ, 340, c332. https://doi.org/10.1136/bmj.c332
- Devereaux, P. J., Manns, B. J., Ghali, W. A., Quan, H., Lacchetti, C., Montori, V. M., Bhandari, M., & Guyatt, G. H. (2001). Physician interpretations and textbook definitions of blinding terminology in randomized controlled trials. JAMA, 285(15), 2000–2003. https://doi.org/10.1001/jama.285.15.2000
- Schulz, K. F., & Grimes, D. A. (2002). Blinding in randomised trials: hiding who got what. The Lancet, 359(9307), 696–700. https://doi.org/10.1016/S0140-6736(02)07816-9
- Hróbjartsson, A., & Gøtzsche, P. C. (2001). Is the placebo powerless? An analysis of clinical trials comparing placebo with no treatment. New England Journal of Medicine, 344(21), 1594–1602. https://doi.org/10.1056/NEJM200105243442106
- World Medical Association. (2024). Declaration of Helsinki: Ethical principles for medical research involving human participants. https://www.wma.net/policies-post/wma-declaration-of-helsinki/
- Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing. Cambridge University Press.
- Lind, J. (1753). A Treatise of the Scurvy. Sands, Murray and Cochran.