The placebo effect
The placebo effect is the change measured in people who received something inert while believing it was active. Note the wording: the change measured, not the change caused. That gap between what a number records and what a number is assumed to mean is the whole subject, and it is the reason this page belongs to methods rather than to health.
The word is Latin for "I shall please" and travels almost unchanged between languages, appearing as efecto placebo in Spanish and effet placebo in French. It entered English usage in the late 1700s to describe something given to satisfy rather than to act, and the modern research meaning kept the shape of that older sense: a thing offered in the position where the real thing would go.
The definition used in psychology, and what it quietly leaves out
Psychology courses, including AP Psychology, usually define the placebo effect as an improvement produced by a person's expectation rather than by any active ingredient. That definition is serviceable for an exam and misleading in practice, because it names a cause. What actually gets measured in a single group of people who took an inert substance is a difference between two points in time, and expectation is only one of the candidates for having produced it.
The confusion has a documented origin. Henry Beecher published a paper in the Journal of the American Medical Association in 1955 titled The Powerful Placebo, pooling 15 studies and reporting that roughly a third of people responded to inert preparations. That figure entered textbooks as evidence of a mind over body effect. In 1997, Gunver Kienle and Helmut Kiene reanalyzed Beecher's own source studies in the Journal of Clinical Epidemiology and found that the reported improvements were explainable without any placebo effect at all, because none of the original trials had an untreated comparison group. The number was real. The interpretation was not.
The four things hiding inside a measured placebo response
A measured placebo response contains at least four distinct components, and only the last one is what people mean when they say the placebo effect. Separating them is the single most useful thing this page can hand you.
| Component | What it actually is | Why it looks like a placebo effect |
|---|---|---|
| Natural course | The thing was going to change on its own over that period | The inert item was taken during the change, so it gets the credit |
| Regression to the mean | Extreme measurements drift back toward average on remeasurement | People join studies at their worst point, so the next reading is almost always milder |
| Reporting and demand effects | People give a more favorable answer to someone who has helped them | The score moves while nothing underlying it does |
| Expectation | Belief that something is working changes perception and, for some outcomes, physiology | This one is the placebo effect proper, and it is the smallest of the four |
Francis Galton described the second of these in 1886, working on human height: unusually tall parents have children closer to average, not because of any influence but because any extreme value is part signal and part noise, and noise does not repeat. Anything selected for being extreme will look better next time. That single statistical fact accounts for a large share of every testimonial ever written about anything.
The most direct evidence on the size of what remains comes from trials with three arms rather than two. Asbjorn Hrobjartsson and Peter Gotzsche pooled such trials in the New England Journal of Medicine in 2001, comparing groups given an inert preparation against groups given nothing at all. Where an outcome was objective and measured by an instrument, they found little sign of a placebo effect. Where an outcome was continuous and self reported, particularly pain, they found modest effects. Belief moves reports reliably. It moves instruments much less.
Why the placebo effect forces a control group into any honest test
The placebo effect forces a control group into any honest test because the four components above are inseparable from the outside. Look at a single group over time and you cannot tell which one produced the change, no matter how carefully you measure, how many people you enroll or how sincere they are. The information simply is not in the data.
Run the same period of time past a second group that received something inert and indistinguishable, and the arithmetic changes completely. Natural course, regression to the mean and reporting effects apply to both groups equally, so subtracting one from the other cancels them. Whatever survives the subtraction is attributable to the intervention. This is why control groups and blinding are not bureaucratic caution but the actual instrument of measurement, and it is why the scientific method treats an uncontrolled before and after comparison as a source of hypotheses rather than of conclusions.
Blinding is the other half of the same problem. Someone who knows they received the inert version will not report like someone who thinks they received the real one, and someone assessing them who knows the assignment will not score them the same way. Expectation is not confined to the participant.
What framing changes, and what it does not
Framing changes the size of the reported placebo response without changing anything about the inert item itself. The ritual around a thing carries information: how confidently it was presented, how much attention came with it, how elaborate the procedure was, what the person was told to expect and how soon. Two groups given identical inert items with different scripts produce different numbers, which is a genuine finding about people and a serious nuisance for anyone trying to measure something else.
What framing does not change is the underlying state of anything the participant cannot perceive. This is the boundary that gets crossed in casual writing about the placebo effect, and it is worth stating flatly: an effect on how something is reported is not evidence of an effect on what is being reported about. Confirmation bias then does the rest of the work, because a person who expects improvement notices and remembers the hours that fit.
Placebo, Barnum and halo are three different effects
The placebo effect is regularly confused with two neighbors that also involve a person filling in a blank with their own expectations, and the three have different mechanisms.
- Placebo effect: expecting a change from something inert alters how you perceive and report your own state. Studied with inert comparison arms.
- Barnum effect: accepting a vague, universally applicable description as a specific insight about yourself. Bertram Forer demonstrated it in 1948 with a single personality sketch handed to 39 students who each believed it was theirs, and Paul Meehl gave it the Barnum name in 1956.
- Halo effect: letting one favorable judgment color unrelated judgments about the same person or object. Edward Thorndike documented it in 1920 in officer ratings, where physical bearing pulled ratings of unrelated abilities along with it.
All three are reasons a first impression of effectiveness is worth very little on its own. Only the first is about an inert intervention.
What the placebo effect does not license
The placebo effect does not license the conclusion that belief is a mechanism of change, and this page recommends nothing about anything. Three specific inferences do not follow from the evidence: that a reported improvement in an uncontrolled group demonstrates the power of the mind, that a large placebo response makes an inert item useful, and that because expectation influences reports, the distinction between an active and an inert item matters less than it appears. The literature runs the other way. The better controlled the outcome, the smaller the placebo response tends to be. Claims built on the opposite reading, which usually arrive with the word natural and without a comparison group, belong to the family of arguments described under pseudoscience.
The test to run on any before and after story
Anyone can run this test on the next testimonial they read, and it takes four questions. Ask them in order, because the first one usually settles it.
- Compared with what? If the answer is "with how I felt before", there was no control group and no measurement of the effect has occurred.
- When did they start? If they started at their worst point, regression to the mean predicts improvement by itself.
- Who did the measuring, and did they know what the person had taken?
- Is the outcome something an instrument recorded, or something a person said afterward to someone who wanted good news?
The honest position after those four questions is often that you cannot tell. That is not a failure of the test. It is the test working, and it is exactly the position the person telling the story never reaches.