A control group is the set of participants who do not receive the change being tested, so researchers have a baseline to compare against. Without one, any improvement could come from timing, mood, the placebo effect, or plain chance.
That comparison is why control groups sit at the heart of drug trials, A/B tests, and survey experiments. A reworded question, a redesigned email, or a wellness program only proves its worth when a similar group went without it.
In this guide, you’ll learn what a control group is, the seven main types, how to size and design one, and how to read the results.
What is a control group in research?
A control group is a set of participants in an experiment who do not receive the treatment, change, or intervention being tested. Researchers compare their results with those of the experimental group to see what the change actually did.
Think of it as the “what would have happened anyway” group. If both groups improve by the same amount, the change did nothing. If only the experimental group improves, the change likely worked.
The idea reaches far beyond medicine. A survey researcher might show the control group the original question and the experimental group a reworded one. A marketer might send the usual email to the control group and a new design to everyone else. In each case, the only planned difference between the groups is the change itself. That is the core logic of experimental research.
A valid control group needs three things:
- Similarity: It matches the experimental group on age, behavior, and other traits that affect the outcome.
- Separation: Its members do not receive the change, directly or indirectly.
- Equal treatment: Everything else, from timing to instructions, stays the same.
What is the difference between a control group and an experimental group?
The control group does not get the change, and the experimental group does. Everything else about the two groups should match, so any gap in results points back to the change.
| Feature | Control group | Experimental group |
|---|---|---|
| Receives the change? | No | Yes |
| Main purpose | Shows the baseline outcome | Shows the outcome with the change |
| What researchers measure | Natural change, placebo response, outside influences | Total effect of the change |
| Survey example | Answers the original question wording | Answers the reworded question |
| Drug trial example | Takes a placebo or standard care | Takes the new drug |
Some fields call the experimental group the treatment group. Both names describe the same thing.
Control group vs similar terms: What is the difference?
Several terms sound alike and get mixed up. Keeping them straight prevents confusion when you read studies or write your own design.
- Treatment group: Another name for the experimental group, the people who receive the change.
- Comparison group: A group used for comparison that was not randomly assigned. It is common in quasi-experimental studies, where random assignment is not possible.
- Control variable: A factor you hold constant for everyone, such as survey length or time of day. It is not a group of people.
- Controlled experiment: A study that includes a control group and changes one thing at a time.
- Baseline: A measurement taken before the change, often from the same people. A baseline is not a control group, because it cannot tell you what would have happened over time.
- Placebo group: One specific kind of control group that receives an inactive substitute.
Why do control groups matter in research?
Control groups let you separate the effect of a change from everything else that moves at the same time. They are the main tool researchers use to claim cause and effect.
Here is what a control group protects you from:
- The placebo effect: People often improve because they expect to. In a classic 1955 review, Henry Beecher reported that placebos helped about 35% of patients across 15 studies, although later work treats that figure as an upper limit.
- Natural change: Customers recover from a bad experience, and symptoms fade. A control group shows how much would have changed anyway.
- Outside events: A holiday, a news story, or a competitor’s sale can shift results for everyone.
- Hidden bias: Random assignment spreads unknown differences evenly across groups, which reduces participant bias.
The result is stronger internal validity, which means you can trust that the change, not something else, caused the outcome.
What are the types of control groups?
There are seven common types of control groups. The US Food and Drug Administration’s ICH E10 guidance names five for clinical trials: placebo, no-treatment, dose-response, active, and external (historical). Behavioral and business research also uses waitlist and attention controls.
Placebo control group
A placebo control group receives an inactive version of the treatment that looks real. It separates the treatment’s effect from participants’ expectations. In a drug trial, that is a sugar pill. In a survey test, it could be a neutral screen that looks like the new feature but changes nothing.
- Best for: Studies where expectations could shape the outcome.
- Watch out for: Ethical limits when an effective treatment already exists.
Active (positive) control group
An active control group receives an existing treatment already known to work. Researchers use it when withholding treatment would be unethical. They also use it to confirm that the study can detect a real effect at all.
A survey team testing a new onboarding questionnaire could compare it with the current format that already earns a strong survey response rate. If the new version cannot beat or match it, the design needs work.
No-treatment control group
A no-treatment control group receives nothing new and keeps its normal experience. It shows the natural course of events and gives the cleanest baseline.
- Best for: Low-risk changes such as a new survey layout or a redesigned page.
- Terminology note: In lab science, a “negative control” is a condition expected to show no effect, used to confirm the setup works. Many business articles use the term as a synonym for no-treatment, so check how a source defines it.
Dose-response control group
A dose-response design uses several groups that receive different amounts of the same change. It shows whether more of the change produces more of the effect. A researcher testing survey incentives might offer no reward to one group, $2 to another, and $5 to a third. If response rates rise with each step, the incentive works, and you learn where returns level off.
Waitlist control group
A waitlist control group receives the intervention after the study ends. This is common in training and therapy research, because nobody is permanently denied the benefit.
- Best for: Programs that people want, such as employee training.
- Watch out for: Inflated results. An analysis of 333 psychotherapy trials found larger average effects against waitlist controls (g = 0.95) than against care-as-usual controls (g = 0.63).
Historical (external) control group
A historical control group uses data from earlier studies or past records instead of recruiting new participants. It saves time and cost. A team might compare this quarter’s complaint-resolution time with last year’s records after launching a new support process.
The risk is that conditions change over time. Staff, customers, and seasons differ, so differences may not come from your change.
Attention control group
An attention control group gets the same time and contact as the experimental group but without the active ingredient. It rules out the possibility that simply being noticed drives the result. This effect is known as the Hawthorne effect, where people change behavior because they know someone is watching.
- Example: One set of employees attends a stress-management workshop. The control set attends a general company talk of the same length.
- Best for: Studies of coaching, training, and counseling.
How do you choose the right type of control group?
Choose the type that answers your question while keeping participants safe and your data clean. Start with ethics, then check cost, time, and how much confidence you need.
| Your situation | Best control group | Why |
|---|---|---|
| Testing a low-risk survey or page change | No-treatment | Fastest and cleanest baseline |
| An effective treatment already exists | Active | Withholding it would be unfair or unsafe |
| Expectations could sway results | Placebo | Separates belief from real effect |
| You need to find the right amount | Dose-response | Shows where more stops helping |
| Everyone should eventually benefit | Waitlist | Delays the benefit instead of denying it |
| You cannot recruit a new group | Historical | Reuses existing records, with caution |
When two options fit, pick the one with fewer assumptions. A concurrent, randomized control group is usually more trustworthy than a historical one.
What are real-world examples of control groups?
Control groups appear in medicine, marketing, customer research, and HR. These five examples show how the same logic adapts to each setting.
Medicine: The Diabetes Prevention Program
The Diabetes Prevention Program tested ways to prevent type 2 diabetes. It enrolled 3,234 people at 27 US clinical centers between 1996 and 2001.
- Lifestyle group: Intensive diet and exercise coaching. Risk fell 58% against placebo after about three years.
- Metformin group: Medication twice daily. Risk fell 31% against placebo.
- Placebo control group: A pill with no active medicine, plus standard diet and activity advice.
Survey research: Testing a reworded question
A research team suspects a question is confusing. It randomly splits respondents. The control group sees the original wording, and the experimental group sees the revision. Everything else stays identical.
The team then compares completion rate, time spent, skipped answers, and the spread of responses. If the revised version produces fewer skips and more consistent answers, the rewording helped. This approach is a practical way to improve survey research quality.
Marketing and product testing
A retailer redesigns its promotional email. Half the list gets the old email (control) and half gets the new one (experimental). The same logic works for product testing, where the control group keeps using the current app and the experimental group tries the update.
- Illustrative result: Control open rate 20%, experimental open rate 24%.
- Reading it: A 4-point gap, or a 20% relative lift, assuming a large enough sample.
Customer experience: holding out a loyalty program
A brand launches a loyalty program but keeps a randomly selected set of similar customers out of the invitation list. After six months, it compares repeat purchase rates and satisfaction scores between the two groups. Comparing members with non-members who chose not to join would be misleading, because joiners are already more engaged. The random holdout avoids that trap.
Employee experience: A wellness program
An HR team offers a new wellness program to one set of teams and keeps similar teams on their usual routine.
- Measure: Engagement scores, sick days, and pulse survey results.
- Timing: Collect them before launch and again after three and six months.
- Reading it: Compare the change in each group, not just the final scores.
Step-by-step guide to designing a control group
A good control group design is mostly decided before any data arrives. Follow these steps in order.
- State one hypothesis and one main metric.
For example: “The revised question will raise completion rate.” - Define who is eligible.
Set inclusion and exclusion rules. Exclude people who took a similar survey recently, since they may remember the questions. - Pick the control group type.
Use the decision table above. - Assign participants at random.
Do not let people or researchers choose groups. - Keep conditions identical.
Match instructions, question order, timing, device, and platform across groups. - Set sample size and duration up front.
Decide when the test ends before you see any results. - Plan the analysis and dropouts.
Name the statistical test and recruit extra people to cover attrition.
How do you randomly assign participants to a control group?
Random assignment gives every participant an equal chance of landing in either group. It is the single best defense against bias you cannot see.
Simple random assignment
Simple random assignment works like flipping a coin for each person, or using a random number generator. It works well with large samples. With small ones, chance can produce lopsided groups, such as 14 people in one and 26 in the other.
- Best for: Samples of several hundred or more.
- Weakness: Small samples can end up unbalanced.
Stratified random assignment
Stratified assignment first sorts people into subgroups, called strata, such as age bands or customer tiers. It then randomizes within each stratum. This guarantees that both groups include the same share of each subgroup, which matters when a trait strongly affects the outcome.
Blocked random assignment
Blocked assignment keeps group sizes equal as people arrive. In a block of four, two people go to control and two go to the experimental group, in random order. The pattern repeats for each new block.
- Best for: Studies where participants join over time.
- Benefit: Group sizes never drift far apart.
Matched pairs
Matching pairs up participants with similar traits, such as age and prior purchases, then puts one of each pair in each group. It helps when samples are small or when full randomization is not possible. Matching only balances traits you measured, so it is a backup, not a replacement for randomization.
How do blinding and consistent conditions protect your results?
Blinding means hiding who is in which group, so expectations cannot change behavior or judgment. It protects against bias from both participants and researchers.
- Single-blind: Participants do not know their group.
- Double-blind: Neither participants nor the people collecting or analyzing data know the groups.
The effect is measurable. Unblinded trials exaggerated subjective outcomes by around 23%.
In a survey experiment, respondents should never see the other version. Analysts should work from coded labels, such as “Group A” and “Group B,” until the analysis is done. Also keep the environment the same: the same invitation text, send time, reminders, and incentives for both groups.
How large should a control group be?
A control group needs enough people to detect the effect you care about. Too few, and a real change looks like noise. One unusual answer among 20 respondents can swing the whole result.
The required size depends on the effect size, which is how big a difference you expect. These figures assume 80% power (an 80% chance of detecting a real effect) and a 5% significance level, for a standard two-group comparison of averages.
| Expected effect | Cohen’s d | People per group |
|---|---|---|
| Small | 0.2 | About 394 |
| Medium | 0.5 | About 64 |
| Large | 0.8 | About 26 |
Proportions need more people. Detecting a rise from a 10% to a 15% click rate takes roughly 690 people per group.
Then add a cushion for dropouts. If you expect 10% attrition and need 200 per group, recruit about 222 (200 divided by 0.9). Use a sample size calculator to check your own numbers.
How do you measure and analyze control group results?
Compare the change in the experimental group with the change in the control group, then test whether the gap is bigger than chance. Look at both statistical and practical significance.
Take a satisfaction survey as an example. Both groups score 70 before the test. After the test, the control group scores 71 and the experimental group scores 78. The experimental group gained 8 points and the control group gained 1, so the estimated effect of the change is 7 points. Analysts call this a difference-in-differences.
- Choose the right test: Use a t-test for averages and a chi-square test for percentages.
- Check the p-value: A p-value below 0.05 means a gap this large would be unlikely if the change did nothing.
- Report effect size and a confidence interval: They show how big the effect is and how certain you are.
- Ask if it matters: A 0.2-point gain can be statistically significant and still not worth the cost.
What are the most common control group mistakes?
Most failed experiments break a few basic rules. Watch for these seven.
- Letting people pick their group: Self-selection brings in hidden differences.
- Contamination: Control participants hear about or see the change, which shrinks the gap.
- Changing the test midway: Edits mid-test make results impossible to read.
- Stopping early: Ending a test when results look good inflates false positives.
- Testing several changes at once: You cannot tell which one caused the effect.
- Ignoring uneven dropout: If more people leave one group, the remaining groups no longer match.
- Skipping seasonality: A holiday week can distort a short test.
What are the limitations and ethical issues of control groups?
Control groups raise ethical questions and practical limits. Knowing both helps you decide when and how to use one.
Ethical concerns
Withholding a helpful treatment from people who could benefit is the main concern. In the US, institutional review boards (IRBs) review human-subject research and require informed consent. Active, waitlist, and dose-response designs let researchers avoid leaving anyone without care. Surveys that only test wording carry little risk, but honesty and consent still apply.
- Rule of thumb: If the change could help or harm people, choose an active, waitlist, or dose-response control.
Practical limits
- Cost and time: Two groups mean more recruiting and more analysis.
- Small samples: Niche audiences make groups hard to fill.
- Narrow results: Findings from one population may not hold for another.
- Outside influences: Mood, timing, and events can still affect responses.
What can you use when a control group is not possible?
When you cannot randomize or withhold a change, other designs can still estimate impact. Each trades some certainty for feasibility.
| Alternative | How it works | Main risk |
|---|---|---|
| Quasi-experimental design | Compares existing groups without random assignment | Groups differ in hidden ways |
| Before-and-after (pre-post) | Measures the same people before and after the change | Time and events can explain the shift |
| Matched comparison group | Pairs similar people who did and did not get the change | Only measured traits are balanced |
| Synthetic control | Builds a weighted “stand-in” group from several similar units | Needs rich historical data |
How does QuestionPro help you run control group studies?
QuestionPro lets you split respondents into randomly assigned groups inside a single survey. This makes the control and experimental versions a built-in feature of survey design instead of a manual process.
The A/B testing workflow uses the Block Randomizer, which is part of the Research Edition:
- Open your survey and choose Tools, then Block Randomizer.
- Add a randomizer section.
- Drag each version of your survey content, the control and the variant, into that section.
- Set how many blocks each respondent sees, then save.
- Export your data with Include Randomization Data turned on to see which version each person received.
Because each respondent sees only one version, they cannot compare conditions. That protects the blinding described earlier. The QuestionPro Market Research Software brings survey design, sampling, and analysis into one place for teams running these tests.
Why a control group makes research believable
A result only means something when you know what would have happened without the change. A control group supplies that missing comparison, and everything else in good design exists to protect it: random assignment, blinding, adequate sample size, and honest analysis.
The best habit is to decide the comparison before you launch. Name the control, the metric, and the stopping point first. Then let the data answer the question rather than the other way around.
Frequently Asked Questions (FAQs)
Yes. Studies often use several control groups, such as a placebo group and an active-treatment group. Each answers a different question. The placebo arm shows the effect of expectation, while the active arm shows whether the new option beats current practice.
In the US, the FDA expects evidence from adequate and well-controlled studies. Its ICH E10 guidance explains which control groups suit different trials. Ethical rules still apply, so an active control often replaces a placebo when effective treatments exist.
A holdout group is a randomly chosen set of customers deliberately kept out of a campaign or program. It works as a control group for ongoing marketing. Comparing holdouts with exposed customers shows the campaign’s true incremental impact on sales or retention.
Run the comparison long enough to cover at least one full business cycle, often one to two weeks for web and email tests. Set the end date before launch. Stopping early when results look good raises the risk of false positives.
It happens often. Placebo response, natural recovery, or seasonal change can lift both groups. The effect of your change is the extra improvement in the experimental group, so a rising control group does not mean the test failed.
