Placebo vs control group: a kind, not a synonym
By Jude Wallis · Published
A control group is the group that supplies the baseline for comparison, whatever it receives. A placebo group is one kind of control group, given a convincing fake so the experience of being treated is present in every arm. Every placebo group is a control group; the reverse is not true.
AP Statistics: Unit 1 (topics 1.13 Experimental Design). Experimental design is topic 1.13, the last topic of Unit 1 (Exploring One-Variable Data and Collecting Data) in the Fall 2026 AP Statistics course, and Unit 1 is 20% to 30% of the multiple-choice section. Collecting data is Practice 2 on the exam, weighted 20% to 30%, so design vocabulary is worth naming precisely.
Placebo vs control group: the short answer
The two words are not on the same level, and that is the whole confusion. One names a role in the design. The other names what the group is handed.
A control group is the group that exists to supply the baseline the treatments are measured against. Its defining feature is its job, not its contents. It might receive an inactive pill, it might receive the treatment already in standard use, it might receive nothing at all and simply be measured.
A placebo group is a control group that receives a convincing fake: an inactive pill identical to the real one, a saline injection, a sham procedure. That is one way of filling the control role, chosen when the act of being treated could move the response on its own.
So the relationship is containment, not equivalence. Every placebo group is a control group, because the fake is given for comparison and nothing else. Most control groups are not placebo groups, because most experiments either cannot build a convincing fake or should not use one.
What a control group is
The control group is what the treatment groups get compared against. Everything about it is handled the same way as the treatment groups except the one factor under test: same room, same schedule, same measuring instrument, and the same random assignment that fills every other group. It is not the pile of units left over after the treatment groups are filled.
Two sentences to stop writing. The first is that the control group is the group that gets nothing. Often it receives the current standard treatment instead, and where an effective treatment already exists that is both the ethical choice and the more useful comparison. The second is that the control group is what earns a cause-and-effect claim. Comparison on its own does not do that; random assignment is what makes the groups alike beforehand, which is the line drawn in experiment vs survey.
Not every experiment has one. Three doses of the same drug compared against each other, or two teaching methods already in use, satisfy the comparison principle without a control group. For the full set of design elements, see how to design an experiment.
What a placebo group is
A placebo is an inactive version of the treatment, and the point is that it is given rather than withheld. A patient who swallows an identical-looking pill has had the whole experience of being treated: the appointment, the instruction, the expectation that something will help. A patient sent home with nothing has not.
That matters because expectation moves responses. Pain scores, energy ratings, and symptom reports all shift when a subject believes treatment is under way. If one arm carries that shift and the other does not, the gap between the arms mixes the active ingredient with the effect of believing. Handing the control arm a placebo puts the belief into every arm, where it cancels in the comparison, and what survives is the ingredient.
The placebo effect has an exact definition, and it is a difference between two averages: the average response to a placebo minus the average response to no treatment at all. It is a quantity you can compute, not a figure of speech, and it is why a placebo group is usually a sharper comparison than an untreated group.
A placebo also makes blinding the subject possible, since there is nothing visible that separates the arms. It does not perform the blinding by itself. If the bottles carry different labels, or the nurse handing them out knows which is which, somebody is not blind, a separation worked through in double-blind.
The differences side by side
| Feature | Control group | Placebo group |
|---|---|---|
| What the word names | The role: the baseline for comparison | The contents: a convincing inactive treatment |
| Receives | No treatment, a placebo, or the current standard | An inactive lookalike of the treatment |
| Relationship | The general category | One kind of control group |
| Works on | Any experimental unit, including plants and machines | Only units that can form expectations |
| Purpose | Give the treatment something to be measured against | Put the experience of being treated into every arm |
| Ruled out when | When every group under test is an active treatment, as in a dose comparison | An effective treatment exists, or no convincing fake does |
Read the first row and the rest follows. Control group answers what this group is for. Placebo answers what is in the bottle.
The three kinds of control group, and when each fits
A control group gets filled in one of three ways, and the question you are asking decides which.
Placebo control. The group receives an inactive lookalike. Use it when the units are people who can form expectations, a convincing fake exists, and no established treatment is being withheld to make room for it. This is the usual choice for a first trial of a new drug in a condition with nothing effective already on the shelf.
Standard-treatment control, also called an active control. The group receives the treatment already in use. Reach for it whenever an effective treatment exists. Withholding a working treatment to hand someone a sugar pill needs a justification a serious condition will not supply, and it also answers the wrong question: what a prescriber needs to know is whether the new treatment beats the current one, not whether it beats nothing. Sham surgery raises the same objection more sharply than a pill does, which is why surgical trials usually compare against the current procedure. Note the cost of this choice. The comparison is now new against standard, so a small difference no longer means the new treatment does nothing.
No-treatment control. The group receives nothing, though it is still measured on the same schedule under the same conditions. Use it when a placebo is impossible or pointless. There is no inactive version of a twelve-week exercise program or a smaller class size, and a plant or a machine has no expectations to manage, which is why agricultural and engineering trials do not use placebos.
A three-arm design uses two of these at once. That is what measuring the placebo effect requires: a placebo arm and a no-treatment arm in the same study, since the quantity is defined as the placebo mean minus the no-treatment mean. Add the treatment under test and you have the third arm.
The classic mix-up and how to avoid it
The mix-up is treating the two words as interchangeable, and it shows up in two sentences that both cost credit.
The first is "the control group gets a placebo." Sometimes it does. Written as an unqualified rule it produces a placebo in a plant trial, or a sugar pill handed to patients whose working medication should not have been stopped.
The second runs the other way: "there is no placebo here, so there is no control group." The comparison group is a control group whatever it receives. Sixty laptops split between new battery firmware and the shipping firmware have a control group and no placebo anywhere in the design.
Two checks settle it. Ask what the group is for: if it exists to be the baseline, it is a control group, full stop. Then ask what it was handed: only if the answer is a convincing inactive version of the treatment is it also a placebo group. Role first, contents second, and the containment falls out. When you write a design description, name every treatment including the control and say exactly what the control receives, which is on the checklist in how to describe a completely randomized design. To rehearse it, use experimental design practice.
One drug, three control groups, three different effect estimates
A trial of a new pain drug randomly assigns 240 patients to four equal groups. Group A takes the new drug, group B takes an identical inactive pill, group C takes the current standard drug, and group D takes nothing but is measured on the same schedule in the same clinic. Mean pain reduction, in points on a 0 to 10 scale, comes out 4.2 for A, 2.6 for B, 3.5 for C, and 1.4 for D. Say which groups are control groups and which are placebo groups, compute the estimated effect of the new drug against each control, and compute the placebo effect.
Check the group sizes. Four equal groups out of 240 patients gives patients per group, and , so every patient is assigned.
Sort the groups by role. Group A is the treatment group under test. Groups B, C, and D each exist to supply a baseline that A is measured against, so all three are control groups.
Sort the same groups by contents. Only group B receives a convincing inactive version of the drug, so group B is the one placebo group. Groups C and D are control groups that are not placebo groups.
Estimated effect against the placebo control: points.
Estimated effect against the standard-treatment control: points.
Estimated effect against the no-treatment control: points.
Placebo effect, defined as the average response to a placebo minus the average response to no treatment: points.
Check the design against the ethics point in the section above. Groups B and D go without the standard drug that group C receives, which a trial can only defend when the condition is mild and the trial short enough that going untreated for its duration costs the patient nothing lasting. Where that is not true, the design drops B and D and compares A against C alone.
Read the spread. Three different baselines give 0.7, 1.6, and 2.8 points for the same drug, a range of points, which is larger than two of the three estimates themselves. Which number you report is decided by which baseline you chose, so a design has to say what the control received, not merely that it had one.
See how the largest estimate breaks apart: , so the no-treatment comparison is the drug's own effect plus the placebo effect. If an effective drug is already on the market, group C is the honest comparison and 0.7 points is what a prescriber needs. If nothing effective exists, group B isolates the active ingredient and 1.6 is the answer. The 2.8 is the one to distrust, because it credits the drug with 1.2 points that came from being treated at all.
Groups B, C, and D are all control groups, and only group B, the identical inactive pill, is a placebo group. The estimated effect of the new drug is points against the placebo, points against the current standard, and points against no treatment, while the placebo effect is points. The 2.8 splits exactly into the drug's own 1.6 and the placebo's 1.2, which is why comparing against no treatment overstates the active ingredient.
Frequently asked questions
Is every control group a placebo group?
No, and that is the direction people get wrong. A control group is defined by its job, supplying the baseline, so it can receive nothing, a placebo, or the current standard treatment, and only the second of those makes it a placebo group. The containment runs the other way: a group given a placebo is given it for comparison, so it is always a control group.
When would a placebo be unethical?
Most often when an effective treatment already exists and the placebo arm would have to go without it. Giving a patient an inactive pill instead of medication known to work withholds care in order to answer a question the standard treatment could answer better. In that case the control group receives the current standard, and the trial compares new against current rather than new against nothing. Sham surgery raises the same objection more sharply.
Can an experiment have a control group with no placebo?
Yes, and most do. Any trial on units that cannot form an expectation qualifies: a field plot given a new fertilizer against one given the fertilizer already in use, or the laptop firmware comparison in the section above. The same holds wherever no convincing inactive version exists, such as a twelve-week exercise program or a smaller class size.
Does using a placebo make a study double-blind?
No. A placebo makes it possible to keep subjects unaware of their assignment; it does not do the keeping. If the bottles are labeled differently, or the person handing them out and recording the response knows which is which, the study is not double-blind. Blinding is about who knows, and a placebo is about what a group receives.