The placebo effect refers to changes in symptoms or physiological processes attributable to the expectations, learned associations, and social context surrounding treatment, rather than to the treatment’s specific therapeutic ingredients. A placebo may be an inactive tablet, a simulated procedure, or another intervention designed to resemble treatment without containing the component under investigation. Placebo effects can also accompany effective medicines and procedures: the biological action of a drug and the effects of knowing that treatment is being given are not mutually exclusive.
The concept is important to medicine because it connects the experience of illness with learning, communication, and measurable biological regulation. It is also central to interpreting clinical research. Improvement among participants receiving a placebo is not, by itself, evidence of a placebo effect: symptoms may improve naturally, fluctuate, or be reported differently during a study. Modern research therefore distinguishes placebo response, meaning the observed change after placebo administration, from a causal placebo effect, meaning the change attributable to the treatment context itself. The strongest evidence concerns particular outcomes, especially pain and other patient-reported symptoms, rather than a general ability to reverse disease.
Historical development and changing definitions
The word placebo comes from Latin and means “I shall please.” It entered medical usage before the development of modern drug testing, when it could describe treatments administered primarily to satisfy patients rather than to act on an identified disease process. Historical placebo use was therefore closely connected with reassurance, professional authority, and uncertainty about the effectiveness of available remedies. This meaning differs from the modern experimental concept, in which a placebo is a tool for separating specific treatment effects from other influences.
Comparisons involving simulated or inactive treatments preceded twentieth-century clinical trials. In 1799, the English physician John Haygarth investigated “Perkins tractors,” metal instruments promoted as treatments for various complaints. Wooden imitations could produce reported improvement, challenging the claim that the original instruments possessed a distinctive therapeutic property. Such demonstrations anticipated placebo-controlled research, although they lacked many features now used to protect against bias, including rigorous randomization and standardized outcome assessment.
The growth of the randomized controlled trial made placebo groups increasingly important during the twentieth century. Henry K. Beecher’s 1955 article “The Powerful Placebo” became particularly influential. Drawing on 15 studies involving 1,082 patients, Beecher reported a frequently quoted figure of approximately 35% for satisfactory relief associated with placebo administration. The article helped establish the placebo as a major medical phenomenon, but its interpretation conflated several sources of improvement. Without suitable untreated comparison groups, the contribution of the placebo context could not be separated from natural recovery and other changes.
A major reassessment came from systematic reviews by Asbjørn Hróbjartsson and Peter Gøtzsche, beginning with a widely discussed publication in 2001. These reviews compared placebo with no-treatment groups rather than simply measuring improvement in placebo arms. They did not support a large, universal placebo benefit across clinical conditions, although effects were found for some outcomes, especially patient-reported continuous measures such as pain. The findings redirected attention from the question of whether placebos “work” in general toward the circumstances, outcomes, and mechanisms for which effects can be demonstrated.
Contemporary terminology remains variable. Some researchers use placebo effect narrowly for experimentally isolated expectation or conditioning effects; others include a broader range of contextual influences, such as therapeutic interaction and treatment ritual. “Contextual treatment effect” emphasizes that these influences are not confined to inactive interventions. However, broader terminology can obscure causal distinctions if every improvement occurring during care is included. A precise account identifies the intervention, the comparison condition, the outcome, and the time interval rather than treating the placebo effect as a single, fixed quantity.
Causal inference and measurement
The central methodological problem is that change after treatment does not establish change because of treatment. A person with an episodic headache, for example, may enter a study when symptoms are unusually severe. Subsequent improvement could reflect the ordinary course of the episode even if the study intervention has no effect. A placebo group controls some influences shared with an active-treatment group, but improvement within that group does not identify which influences caused the change.
One important process is regression to the mean. When participants are selected because a measurement is extreme, later measurements tend, on average, to be less extreme because transient influences and measurement error are unlikely to recur in exactly the same way. Natural history is distinct: an infection may resolve, or a fluctuating condition may enter remission. Additional contributors include changes in daily behavior, use of other treatments, repeated testing, and selective loss of participants whose symptoms do not improve. None is automatically a placebo mechanism.
A three-arm study can distinguish some of these components by randomly assigning participants to active treatment, placebo, or no study treatment. For an outcome expressed as improvement, a simplified estimate is
This contrast estimates the effect of assignment to the placebo condition relative to the chosen control. It does not necessarily isolate a single psychological mechanism. Receiving a placebo may involve extra appointments, a persuasive explanation, physical contact, or the expectation of benefit. Conversely, knowing that treatment has been withheld may cause disappointment. The estimated effect therefore depends on what differs between the groups.
Measurement further complicates interpretation. A patient-reported outcome records an experience such as pain, nausea, or perceived disability. Expectation may genuinely change that experience, but it may also change how participants interpret a scale or what they choose to report. These possibilities cannot always be separated by questionnaires alone. Objective measures, behavioral tasks, blinded assessment, and physiological recording can provide complementary evidence, although no single measurement captures every clinically meaningful aspect of illness.
Randomization strengthens causal inference by balancing prognostic factors on average, but it does not eliminate all bias. Participants generally know whether they receive no treatment, and their knowledge may affect adherence, reporting, or additional care. Small samples, incomplete follow-up, and flexible outcome selection can also exaggerate findings. Accordingly, a statistically detectable placebo effect is not automatically large, durable, or important to daily functioning. Interpretation requires the effect size, uncertainty, outcome relevance, and comparison condition, not merely a statement that the placebo group improved.
Expectations, learning, and interpersonal context
Expectations are beliefs or predictions about what is likely to happen after treatment. They can arise from an explicit statement—such as being told that a procedure is likely to relieve pain—or from prior experience, the appearance of a medical device, or the setting in which care is delivered. Expectations may influence attention, emotional appraisal, and the interpretation of bodily signals. A sensation understood as temporary and controllable may be experienced differently from the same sensation interpreted as evidence of worsening injury.
Classical conditioning provides a partly distinct mechanism. If an active drug repeatedly produces a physiological response while being paired with a distinctive taste, capsule, or setting, the accompanying cue may later elicit part of that response without the full drug intervention. Experimental studies have investigated conditioned changes in pain, hormonal secretion, and immune measures. Such findings do not establish that every drug effect can be reproduced by conditioning: the strength, duration, and biological specificity of the learned response depend on the system and the training procedure.
Learning can also occur through observation. Watching another person appear to benefit from a treatment may alter expectations and subsequent symptom responses. Verbal information and observed experience can reinforce one another, but they can also conflict. A participant told that a cream is analgesic may learn otherwise if repeated testing produces no relief. Experimental placebo analgesia studies sometimes use a conditioning phase in which stimulus intensity is secretly reduced after application of an inactive cream, creating an apparent experience of successful treatment.
The interpersonal setting contributes through communication, credibility, and perceived support. In a 2008 trial involving irritable bowel syndrome, participants receiving sham acupuncture were assigned to interactions that differed in warmth, attention, and expressed confidence. Outcomes were better in the augmented-interaction condition than in the limited-interaction condition. The study illustrated that a treatment encounter can be experimentally varied, although it did not show that one interpersonal style has a uniform effect across diseases or cultures. Time spent with a practitioner can also alter reporting and behavior independently of expectation.
Predictive accounts of perception offer a broader interpretation: the nervous system combines incoming bodily information with prior expectations to construct experience. This approach is influential in placebo research but is not a complete, experimentally settled explanation of every effect. Conscious optimism is neither necessary for all conditioned responses nor sufficient to generate substantial clinical improvement. Individuals also do not divide reliably into permanent “placebo responders” and “nonresponders.” A person’s response can vary with the symptom, previous experience, instructions, and treatment setting, limiting the usefulness of a single personality-based label.
Neural and physiological mechanisms
Placebo analgesia provides some of the clearest evidence that treatment context can engage specific biological pathways. Pain is not a direct readout of tissue damage: it is shaped by sensory input, attention, threat, memory, and descending regulation from the brain. Expectations of relief can recruit systems that alter how nociceptive information is processed. This does not mean that pain is imaginary, or that a person consciously chooses whether to experience it.
A landmark experiment by Jon Levine, Newton Gordon, and Howard Fields, published in 1978, investigated postoperative dental pain. The opioid antagonist naloxone reduced placebo-associated analgesia in participants who had initially responded to placebo. The finding supported a role for endogenous opioid signaling in at least some placebo pain relief. Later work showed that placebo analgesia is mechanistically heterogeneous: opioid blockade does not abolish every experimentally induced placebo effect, and the preceding learning experience can influence which pathways are recruited.
Studies using positron emission tomography and functional magnetic resonance imaging have examined changes in pain-related networks and neurotransmitter activity. Regions implicated include prefrontal and cingulate areas, the periaqueductal gray, and other structures involved in descending pain regulation. Some experiments have also detected modulation of responses within the spinal cord, supporting an influence on relatively early stages of nociceptive processing. Nevertheless, a change in an imaging signal is not itself proof of clinically meaningful pain relief, and activity in a named brain region rarely identifies a unique psychological process.
Research in Parkinson’s disease provides a different example. A 2001 imaging study by Raúl de la Fuente-Fernández and colleagues demonstrated striatal dopamine release associated with placebo administration. Expectations of receiving an effective treatment can therefore engage a neurotransmitter system central to the disease’s motor manifestations. Subsequent experiments have investigated how stated probabilities of receiving active medication influence this response. Such observations concern short-term neurochemical or motor changes; they do not show that placebo treatment prevents the underlying loss of dopaminergic neurons.
Conditioning studies have also examined the immune system and endocrine regulation. Pairing a distinctive drink with an immunosuppressive drug, for example, has produced conditioned changes in selected immune measures in experimental settings. These findings demonstrate communication between learning processes and physiological regulation, not a general capacity to control immunity voluntarily. Effects on laboratory measures require separate evaluation from effects on disease outcomes. Across mechanisms, the relevant question is which pathway changes, by how much, for how long, and with what consequence—not whether there is a single “placebo center” in the brain.
Clinical evidence and the limits of symptom improvement
The clinical evidence is uneven across conditions and outcomes. Placebo effects are most consistently demonstrated for some subjective symptoms, particularly experimentally induced pain and certain forms of clinical pain. Effects have also been investigated in nausea, fatigue, sleep complaints, and gastrointestinal symptoms. Their magnitude varies with the comparison group, the credibility of the intervention, the duration of observation, and the way outcomes are measured. A strong effect in a short laboratory experiment does not establish an equally strong effect during long-term disease management.
Pain research is informative because the sensory experience can change even when the physical stimulus is held constant. However, evidence from experimental heat or pressure pain cannot simply be transferred to all forms of chronic pain. Persistent pain involves differing combinations of tissue pathology, nerve injury, central sensitization, sleep disturbance, and disability. A reduction in reported pain may be meaningful without correcting the underlying lesion, while an unchanged structural finding does not invalidate a genuine change in suffering.
A frequently cited 2011 study in asthma illustrates the distinction between subjective and physiological outcomes. Participants received an active albuterol inhaler, a placebo inhaler, sham acupuncture, or no intervention in a crossover design. The placebo interventions produced substantial reported improvement, but the active bronchodilator produced a much larger improvement in forced expiratory volume in one second, an objective measure of airflow. The experiment showed that perceived breathing relief and reversal of airway obstruction can diverge. It did not establish that subjective symptoms are unimportant; rather, both types of outcome answer different clinical questions.
Similarly, improvement in symptoms associated with cancer does not demonstrate tumor regression, and altered pain does not establish fracture healing or eradication of an infection. Evidence for contextual effects on particular physiological variables cannot be generalized to disease modification. Studies of depression present additional interpretive challenges because rating scales include symptoms sensitive to attention and expectation, illness often fluctuates, and trial participation may provide substantial supportive contact. High improvement rates in placebo arms do not by themselves establish a large causal placebo effect.
Duration is another limitation. Many mechanistic experiments observe responses over minutes or hours, whereas clinical studies commonly assess weeks or months. Expectations can weaken when an intervention repeatedly fails to deliver anticipated benefit, although learned effects may sometimes persist or be renewed. Evidence about lasting changes in disability, healthcare use, and quality of life is more limited than evidence about short-term symptom ratings. Clinical importance therefore depends on the specific endpoint and time frame, rather than on an undifferentiated claim that placebo effects are either powerful or negligible.
Nocebo effects and adverse symptom attribution
The nocebo effect is the counterpart of placebo benefit: negative expectations or learned associations contribute to worse symptoms or adverse experiences. A nocebo response, however, is not synonymous with every adverse event occurring in a placebo group. Participants may develop headaches, nausea, or other complaints because these symptoms occur in everyday life, because their illness is changing, or because another treatment is responsible. A causal nocebo effect requires evidence that the relevant expectation or context increased the adverse outcome.
Experimental studies can manipulate this context directly. An inactive substance described as increasing pain may heighten pain ratings compared with the same substance presented neutrally. Negative information can increase anticipatory anxiety, direct attention toward bodily sensations, and encourage interpretation of ambiguous sensations as harmful. Learning from previous adverse experiences can produce similar effects. Mechanistic pain studies implicate several pathways, including anxiety-related processes and cholecystokinin signaling, but no single pathway explains all nocebo phenomena.
Adverse-event reporting in randomized trials demonstrates why attribution is difficult. Participants assigned to placebo often report symptoms resembling the known adverse effects of the drug under investigation. Trial information, background symptom frequency, expectations, and the underlying condition can all contribute. The observation does not establish that the reported symptoms are fabricated, nor that comparable symptoms during active treatment lack a pharmacological cause. Specific drug toxicity and expectation-related amplification can coexist in the same person.
Blinded crossover studies of statins illustrate this distinction. In the SAMSON trial, published in 2020, 60 participants who had stopped statins because of symptoms underwent periods involving atorvastatin, placebo, and no tablets. Average symptom burden was similar during statin and placebo periods and lower during no-tablet periods. The findings demonstrated a substantial tablet-associated symptom component in that selected sample. They did not show that statins never cause adverse effects, nor did they establish the cause of symptoms in every individual taking them.
Nocebo research creates an ethical tension in risk communication. People require accurate information about potential harms for valid consent, yet the wording and presentation of that information can affect expectations and symptom attention. Positive and negative framing can convey mathematically equivalent frequencies while eliciting different responses. This does not justify withholding material risks. It shows that communication is itself part of the treatment environment and must be studied alongside pharmacological safety. The distinction between adverse events, adverse drug reactions, and expectation-mediated symptoms remains essential to both clinical interpretation and pharmacovigilance.
Placebo controls and the interpretation of clinical trials
In a clinical trial, a placebo control is designed to resemble the active intervention while omitting the component whose efficacy is being tested. Matching may involve appearance, taste, packaging, administration schedule, and the accompanying clinical encounter. Combined with randomization and allocation concealment, masking reduces the possibility that expectations influence treatment delivery, additional care, outcome reporting, or assessment. A placebo is therefore a methodological control, not an assumption that psychological influences are absent.
The active-versus-placebo contrast generally estimates the incremental effect of assignment to the active intervention within that trial’s context. It does not measure the entire benefit of receiving care rather than receiving nothing. Nor is the difference necessarily a chemically pure “drug effect.” Recognizable adverse effects can reveal treatment allocation, and pharmacological actions may interact with expectation. A drug that causes dry mouth, for example, may lead a participant to infer that active treatment was received, altering perceived benefit or the reporting of symptoms.
An active placebo produces a noticeable sensation or minor physiological effect intended to imitate aspects of the experimental treatment without providing its proposed therapeutic action. This approach can improve masking, but it introduces its own problems: the comparator may affect the outcome, cause harm, or still fail to reproduce the active treatment’s experience. Asking participants to guess allocation can provide useful information, although correct guesses may result from genuine improvement rather than from failed masking at the outset.
Procedural trials pose additional difficulties. In surgery, a sham procedure may reproduce preparation, anesthesia, skin incisions, or postoperative care without performing the operative step under investigation. A 2002 trial of arthroscopic surgery for knee osteoarthritis found no advantage of arthroscopic lavage or debridement over placebo surgery on the studied outcomes. Its interpretation concerned those procedures and that population, not surgery generally. Sham-controlled studies must also account for risks introduced solely to preserve masking, including anesthesia and wound-related complications.
Other designs investigate treatment context more directly. In open-versus-hidden administration studies, the same drug is given either with the participant’s awareness or through concealed delivery, allowing the contribution of awareness to be examined while maintaining drug exposure. Balanced-placebo designs vary both what participants receive and what they are told they receive, although deception and recognizable drug effects constrain their use. Factorial designs can separately vary communication and treatment. Each design answers a different causal question, making the exact comparator essential to interpreting the result within evidence-based medicine.
Ethics, open-label placebos, and unresolved research questions
Deceptive placebo administration raises questions of autonomy, trust, and informed consent. Presenting an inactive intervention as an established effective treatment can misrepresent evidence and interfere with a person’s understanding of care. It is especially consequential if placebo use delays diagnosis or replaces effective treatment. These ethical issues are distinct from whether contextual effects are biologically real: demonstration of a mechanism does not by itself justify the conditions under which that mechanism is elicited.
Research ethics imposes further constraints on placebo-controlled trials. The Declaration of Helsinki generally requires comparison with the best proven intervention, while allowing placebo or no intervention under specified conditions, including when no proven intervention exists or when compelling scientific reasons justify it without exposing participants to additional risks of serious or irreversible harm from withholding treatment. Review by an institutional review board considers these issues alongside consent, scientific validity, and risk minimization. An inert comparator is not ethically harmless merely because it lacks an active drug.
Open-label placebos attempt to investigate contextual benefits without concealing that the intervention is inactive. Participants are explicitly told that the pills contain no active therapeutic ingredient, often alongside an explanation of conditioning and mind–body processes. A 2010 randomized trial by Ted Kaptchuk and colleagues enrolled 80 people with irritable bowel syndrome and reported greater short-term symptom improvement with open-label placebo than with no treatment, within a supportive study setting. Subsequent studies have examined conditions including low-back pain and fatigue.
These studies establish a research question rather than a settled clinical conclusion. Participants know their allocation, treatment explanations may themselves create positive expectations, and pill-taking groups may receive a different experience from untreated controls. Small samples, short follow-up, selected volunteers, and reliance on self-reported outcomes limit generalization. Comparisons with credible attention controls and standard care are important for determining what an open-label placebo adds. Disclosure of inert content removes one form of deception but does not eliminate the need for accurate claims about effectiveness.
Several broader questions remain unresolved. Researchers are investigating how to distinguish altered symptom experience from altered reporting, how contextual effects interact with active treatments, and whether conditioned responses can persist under practical clinical conditions. Genetic and brain-based predictors have been proposed, but no general-purpose test reliably identifies placebo responders across disorders and settings. Study registration, adequate sample sizes, prespecified outcomes, and independent replication are particularly important where small experiments and flexible analyses can produce unstable findings. These questions place placebo research at the intersection of learning, neurobiology, trial methodology, and the ethics of therapeutic communication.