Written and maintained by CASRAI Editorial Board
Last updated
Control is important in an experiment because it lets you separate what your treatment did from everything else that could have changed the outcome. Without a control, a before-and-after difference could come from the treatment, from the natural course of events, from what participants expected, or from something else that changed at the same time, and the data cannot tell you which.
Purpose of control in an experiment
The International Council for Harmonisation puts the purpose in one sentence in its guideline ICH E10, Choice of Control Group and Related Issues in Clinical Trials (checked 2026-10-29): control groups have one major purpose, which is to allow discrimination of outcomes caused by the test treatment from outcomes caused by other factors, such as the natural progression of the disease, observer or patient expectations, or other treatment. The guideline is written for clinical trials, but the logic is the general logic of experimental design.
The control group answers a counterfactual question: what would have happened to these participants had they not received the test treatment, or had they received a different treatment? You can never observe that for the same person at the same time, so you estimate it from a comparison group that is as similar as you can make it.
Control group, control variable and experimental control: which one do you mean?
The phrase covers three related ideas that are often blurred.
- Control group. The group that does not receive the experimental manipulation. OpenStax Psychology 2e (section 2.3, Analyzing Findings, checked 2026-10-29) describes the experimental group as the one that gets the manipulation and the control group as the one that does not; in a medication study the control group typically receives a placebo.
- Control variable. A factor other than the one you are testing that you hold constant for everyone, so it cannot explain a difference between groups. See manipulated variable definition for how controlled variables sit alongside the manipulated and measured ones.
- Experimental control (the general sense). Everything in the design that limits alternative explanations, including the comparison group, the held-constant variables and random assignment.
What a control group isolates (ordered list)
ICH E10 section 1.3.1 explains that a placebo control does more than account for the placebo effect. It controls for all potential influences on the actual or apparent course of the disease other than those from the pharmacologic action of the test drug. In order, those influences are:
- Spontaneous change. The natural history of the condition and regression to the mean (extreme measurements tending to be less extreme when repeated). A group measured only before and after treatment cannot rule these out.
- Expectations. What the subjects or the investigators expect to happen.
- The effect of being in a trial. Attention, monitoring and the act of taking part.
- Other therapy. Additional treatment participants receive during the study.
- Subjective elements of assessment. Outcomes that depend on judgement by participants or raters.
A control that is not untreated can still do this job. E10 notes that using a placebo control group does not imply the control group is untreated: in add-on designs both arms receive background therapy and only the added element differs.
Why concurrent control and random assignment matter
E10 distinguishes five types of control group: placebo concurrent control, no-treatment concurrent control, dose-response concurrent control, active (positive) concurrent control, and external (historical) control. The first four are concurrent, meaning the control group is chosen from the same population as the test group and treated in a defined way as part of the same trial. External control is described as usable only in unusual circumstances, because the comparability of people studied at a different time or place is hard to establish.
Assignment matters as much as the comparison group. OpenStax explains that with random assignment every participant has an equal chance of landing in either group, which spreads other characteristics across the groups by chance rather than by anyone’s choice. The related randomization methods entry covers the clinical-trial mechanics. An uncontrolled difference that is tied to the outcome is a confounding variable, and lurking variables are the ones you did not think to measure. OpenStax illustrates this with ice cream sales and crime, where temperature, not ice cream, moves both.
Finally, E10 section 1.5 defines assay sensitivity as the ability of a trial to distinguish an effective treatment from a less effective or ineffective one, and the choice of control group bears directly on whether a trial has it.
A short annotated example
Illustrative composite, not a real study. A researcher wants to know whether a short daily practice routine improves recall scores.
- Without a control. Thirty students do the routine for four weeks and their scores rise. Annotation: scores may have risen because students were simply more familiar with the test, or because the term was nearing an exam. Nothing in the design separates those explanations from the routine.
- With a control group. Sixty students are randomly assigned, thirty to the routine and thirty to carry on as usual, and both groups take the same test at the same times. Annotation: familiarity and exam proximity now affect both groups, so a larger gain in the routine group is much harder to attribute to anything but the routine.
- With control variables. Test version, time of day and scoring rubric are the same for everyone. Annotation: these no longer vary between groups, so they cannot explain a gap.
This is a deliberately simple case. Related ways a design can still be biased are covered in bias in an experiment.
Common mistakes
- Comparing with the past instead of a concurrent group. E10 treats external or historical controls as usable only in unusual circumstances.
- Skipping random assignment. If participants or investigators choose who gets the treatment, the groups can differ in ways that track the outcome.
- Treating a placebo as only about the placebo effect. Per E10 it also controls for natural history, regression to the mean and the effect of being in a trial.
- Mixing up control group and control variable. Holding factors constant does not create a comparison group, and a comparison group does not hold every factor constant.
- Assuming a control means no treatment. Add-on designs give both arms background treatment.
Checklist
- Is there a comparison group that experiences everything except the manipulation?
- Was it drawn from the same population and run at the same time as the treated group?
- Were participants randomly assigned?
- Which other factors are held constant for everyone, and are they written down?
- Does the control type match the question (placebo, no treatment, dose-response, active)?
Discipline differences
Clinical research names its control types explicitly. ICH E10 lists the five control types above. For the clinical design with a placebo arm, see placebo-controlled study design. Psychology teaching texts such as OpenStax use the simpler experimental versus control group framing.
FAQ
Why is a control group necessary?
It shows what happens without the treatment, so the treatment effect can be separated from natural change, expectations and other therapy.
What happens if an experiment has no control?
Any change you observe could have several causes and you cannot say how much the treatment contributed.
Is a control group the same as a control variable?
No. The control group is the comparison group; control variables are factors held constant for all participants.
Does the control group receive nothing?
Not always. It may receive a placebo, an existing treatment, or background therapy that both groups receive.
Sources
- International Council for Harmonisation, ICH E10: Choice of Control Group and Related Issues in Clinical Trials, sections 1.2, 1.3, 1.3.1 and 1.5, checked 2026-10-29.
- OpenStax, Psychology 2e, section 2.3 Analyzing Findings, checked 2026-10-29.
