Placebo Effect: How Expecting to Improve Changes What You Feel

A placebo is inert by definition, so the effect that follows it comes from everything around the pill rather than from the pill. The three main sources are expectation, conditioning and the clinical encounter. Expectation is the belief that improvement is coming, which shifts attention away from symptoms and changes how they are interpreted. Conditioning is the body's learned response to the cues of treatment: a person who has taken effective painkillers many times responds to the sight and taste of a tablet before any drug could act. The encounter is the attention, confidence and reassurance of the clinician, which lower anxiety and with it the symptoms that anxiety amplifies.

Some of this has a measurable biology. Levine, Gordon and Fields showed in 1978 that placebo pain relief after dental surgery was reduced by naloxone, a drug that blocks the body's own opioids, which means the expectation of relief had recruited the same system that morphine uses. Later imaging work found that placebo analgesia changes activity in brain regions involved in pain and in its anticipation. Finniss and colleagues, reviewing the field in The Lancet in 2010, concluded that placebo effects are genuine psychobiological events, strongest for pain and other subjective symptoms, and that there is not one placebo effect but many, each with its own mechanism.

The limits matter as much as the mechanism. In 2001 Hróbjartsson and Gøtzsche pooled 114 trials that included both a placebo group and a no-treatment group and found that placebos had little effect on binary outcomes and on objective measures, with a modest benefit on continuous subjective outcomes and on pain. Their 2010 update with 202 trials reached the same conclusion. A placebo can change how much a knee hurts; it has not been shown to change the cartilage. Claims that placebos shrink tumors or cure infections are not supported.

Much of what gets called a placebo effect in daily life is not one. Symptoms fluctuate, and people seek treatment when they are at their worst, so the next measurement tends to be better whatever was done; that is regression to the mean. Many conditions resolve on their own. In a trial, these are counted in the placebo arm, which is why the placebo arm improves even when the placebo does nothing. Only a comparison with a no-treatment group separates the placebo effect from the passage of time.

Two trials anchor this page. The first is Bruce Moseley and colleagues' trial of arthroscopic surgery for osteoarthritis of the knee, published in the New England Journal of Medicine in 2002. At the Houston Veterans Affairs Medical Center, 180 patients were randomly assigned to arthroscopic debridement, arthroscopic lavage or placebo surgery. The placebo group received anesthesia and three skin incisions while the surgeon simulated the procedure, without inserting an instrument into the joint. Patients and the assessors who measured their outcomes did not know the assignment. Pain and function were followed for 24 months.

At no point in those two years did either surgery group report less pain or better function than the placebo group. The objective measure of function, a timed walk and stair climb, was no better either. With no untreated group, the trial could not say how much of the improvement in all three groups was expectation and how much the natural course of the condition; what it showed was that the procedures added nothing the sham did not.

The second is Ted Kaptchuk and colleagues' trial of open-label placebo in irritable bowel syndrome, published in PLoS ONE in 2010. Eighty patients were assigned either to no treatment or to pills honestly labeled as placebo, with an explanation that placebos had been shown to help through mind-body processes. After three weeks, 59 percent of the placebo group reported adequate relief against 35 percent of the no-treatment group, and the placebo group scored better on global improvement and on symptom severity. The trial was small and short, and the outcomes were self-reported, but it showed that deception is not required. A 2021 meta-analysis by von Wernsdorff and colleagues pooled the open-label trials since and found a moderate benefit on self-reported outcomes across small studies with mixed quality.

The effect, and its look-alikes, appear wherever a treatment is followed by an improvement.

The placebo effect is not an error to be eliminated; it is a reason to compare carefully before concluding that a treatment works.

• Ask for the comparison group. Improvement after a treatment is not evidence for the treatment until it is compared with improvement after a placebo, and ideally with no treatment at all.
• Remember regression to the mean. People seek help when symptoms peak, so the next measurement is usually better. A single before-and-after is not a test.
• Weigh the outcome type. A placebo can change how much something hurts; it has not been shown to change a tumor, a bacterial count or a fracture. Treat claims about objective outcomes with more suspicion than claims about felt ones.
• Keep the relief and question the story. If something helps you feel better, that is real. The question is whether the mechanism claimed for it has been tested, which matters when the treatment is expensive, risky or a substitute for one that has.
• Notice price and brand. Identical products at different prices produce different reported relief. Compare labels, not packaging.

The placebo effect is often used as a catch-all for any improvement without a cause.

• It is not the expectation bias, which is a distortion in what an observer records or measures because of what they expect to find. The placebo effect is a change in what the patient experiences; expectation bias can inflate a trial result with no change in the patient at all.
• It is not the observer-expectancy effect, in which a researcher's expectations leak to participants through cues and change their behavior. Blinding the researcher guards against that; blinding the patient does not remove the placebo effect but keeps expectations alike in both arms, so that it drops out of the comparison. A good trial does both.
• It is not subjective validation, the tendency to accept a vague statement as accurate because it feels personally meaningful. That bias concerns belief in a claim; the placebo effect concerns felt symptoms.
• It is not the nocebo effect, its mirror image, in which expecting harm produces side effects from an inert treatment. The two share mechanisms and differ in direction.