Why does a sugar pill work?

In a 2010 randomised trial, patients with irritable bowel syndrome were given a bottle labelled placebo pills, and told plainly that the pills contained no active medication, while also being told that the placebo effect is powerful and that the body can respond to a placebo the way Pavlov’s dogs responded to a bell.

Over three weeks they improved more than patients on the same schedule of visits with no pills, by 0.79 standard deviations on a self-reported global improvement scale. The authors call it a proof of principle. Eighty patients, three weeks, and the patients could not be blinded, so they could not rule out that part of the difference sits in the reporting rather than in the symptom.

A 2021 meta-analysis of eleven such trials, across back pain, cancer-related fatigue, ADHD, allergic rhinitis, depression, hot flushes and irritable bowel, found 0.72 standard deviations (0.39 to 1.05), falling to 0.49 once the four trials at high risk of bias are dropped. Every primary outcome in it was a self-report.

A 2025 update pooled sixty trials, in patients and in non-clinical samples. It put the effect at 0.47 in patients and, taking every trial together, found it on self-reports (0.39) but not, to any significant degree, on objective measures (0.09).

Whatever is happening is not simply a false belief about a pill, because the belief was corrected in advance.

Placebo analgesia engages the body’s own opioid system. Naloxone, which blocks opioid receptors, reduces both the relief people report and the brain responses that go with it, and abolishes the coupling between cingulate cortex and periaqueductal grey that predicts the effect. That is as direct a demonstration of a physical mechanism as this field offers.

Expectation is not the only ingredient. The review cited here puts placebo effects down to several processes at once, learning and expectation and social cognition, which is why the same pill does different things in different hands.

A Cochrane review of 234 trials that carried a no-treatment arm alongside the placebo arm, 202 of them with usable outcome data, found no important clinical effect from placebo in general. What it found was mainly an effect on what patients report: 0.26 standard deviations on patient-reported outcomes against 0.13 on observer-reported ones, clearest for pain and nausea, and very variable for pain even among the trials at low risk of bias.

Parkinson’s disease is the exception hardest to explain away. A placebo releases dopamine in the striatum of patients whose own supply is failing, which is measurable and is nobody’s report of anything.

The distinction is between changing the course of a disease and changing what it does to a person, and the popular version of the placebo effect ignores it. The reviewers add the caution that goes with their own result: on a self-reported outcome, a placebo effect is hard to separate from biased reporting.

Not every review reads the evidence so narrowly. Wager and Atlas cite controlled studies in chronic pain, depression and Parkinson’s in which placebo moved the measures trials use as primary end points, in some cases as far as accepted drugs do.

Everyone in a trial receives the context, so the measured effect of a drug is the part that exceeds it. Where the placebo arm improves a lot, which includes pain and depression, that subtraction gets difficult and the resulting effect sizes are contested.

Much of that improvement is not the placebo effect. A placebo arm also records the natural course of the condition and regression to the mean. Across 37 trials that added a no-treatment arm, patients given nothing improved a little over half as much as patients given a placebo.

Part of the long argument about how well antidepressants work is really an argument about this.

No accepted account explains the open-label result. Conditioning is the account the trials themselves offer patients. Reinterpretation is another: the instruction contradicts itself, this pill may help and this pill is inert, and that may change what a sensation is taken to mean. And the people who volunteer have often run out of other options, so new hope is hard to exclude.

It is a repeatable effect on what patients report, with no agreed mechanism, which is an unusual and productive position for a result to be in. One laboratory study found open-label relief of experimentally induced pain reversed by naloxone, as ordinary placebo relief is, in the four in ten people who responded at all. The size of the effect is not settled, and on objective measures it has not been shown.