Stanley Milgram

Introduction

Stanley Milgram (15 August 1933 – 20 December 1984) put ordinary people in a room and asked them to hurt someone, and a majority of them did. In the baseline condition of his experiments, participants — told they were taking part in a study of memory and learning — were instructed to administer what they believed were increasingly severe electric shocks to a stranger who was getting the answers wrong. Sixty-five per cent of them went to the maximum on the machine, 450 volts, well past the point labelled "Danger: Severe Shock" and past the point at which the man had stopped responding.

The finding is the most famous in social psychology and the most argued over. What Milgram took himself to have shown is that the capacity for cruelty under orders is not a property of particular people or particular national characters but a property of situations — that the relevant variable is not who a person is but where they have been placed. He was explicit that this was why he ran the study at all: he was a Jew born in the Bronx in 1933, and the question the twentieth century had handed him was how ordinary people come to do what the perpetrators of the Holocaust had done.

He also, almost single-handedly, created modern research ethics. The controversy over what he did to his participants is the reason informed consent, debriefing and institutional review now exist in their present form.

And the archival work of the last fifteen years has substantially revised the story. The 65% figure comes from one condition out of more than twenty; across the full set obedience ran from 0% to 92%; many participants did not believe the shocks were real; the experimenter frequently improvised beyond the scripted prods; and some participants were never properly debriefed. The phenomenon appears to be real. The number in the textbooks is not the whole of it.

Early Life and Education

Milgram was born in the Bronx, New York, to a Jewish family — his father a baker, his parents immigrants from Eastern Europe. He grew up in the shadow of the war and the news from Europe, and the question of obedience was personal to him from early on.

He took his bachelor's degree in political science at Queens College, City University of New York, in 1954. He then applied to Harvard's PhD programme in social psychology and was rejected on the grounds of insufficient psychology preparation; he took courses and was admitted the following year. At Harvard he worked with Gordon Allport, and he served as a research assistant to Solomon Asch, whose conformity experiments were the immediate model for what he would later do — with the crucial difference that Asch's participants were asked to deny what they could see, while Milgram's would be asked to hurt someone.

His dissertation was a cross-national conformity study: he ran Asch's line-judgement procedure in Norway and France, and found Norwegians more conforming than the French. It established the pattern he would use for the rest of his career — take a known laboratory effect and vary one structural feature of the situation to find out what it depends on.

He received his doctorate in 1960.

Academic and Professional Career

Yale, 1960–1963

Milgram joined Yale as an assistant professor in 1960. In April 1961 the trial of Adolf Eichmann opened in Jerusalem, and the phrase that came out of it — "I was only following orders" — supplied the question. Milgram designed the experiment that summer and the first sessions ran in July 1961 in Linsly-Chittenden Hall.

The procedure was a piece of theatre. A participant and a confederate arrived together; by a rigged draw, the participant was assigned the role of "teacher" and the confederate the "learner". The learner was strapped into a chair with an electrode, and the teacher was taken to an adjoining room and shown a shock generator: thirty switches, labelled from 15 to 450 volts, grouped under "Slight Shock", "Moderate Shock", "Strong Shock", "Very Strong Shock", "Intense Shock", "Extreme Intensity Shock", "Danger: Severe Shock", and finally two switches marked only "XXX". The teacher was given a sample 45-volt shock, so the apparatus felt real.

The task was a paired-associate memory test. Each error was to be punished with a shock, increasing one step at a time. The learner — in fact an actor — gave mostly wrong answers. At 75 volts he grunted; at 120 he complained; at 150 he demanded to be released and said his heart was bothering him; at 180 he screamed that he could not stand the pain; at 300 he refused to answer at all; beyond 345 there was silence. The experimenter, in a grey lab coat, responded to hesitation with a fixed sequence of four prods: "Please continue." "The experiment requires that you continue." "It is absolutely essential that you continue." "You have no other choice, you must go on."

Over 1961 and 1962 Milgram ran more than twenty variations — around 636 participants in total — varying proximity, the prestige of the setting, the presence of dissenting peers, and the behaviour of the experimenter. The baseline paper appeared in the Journal of Abnormal and Social Psychology in 1963.

Harvard, and the tenure that did not come

Milgram moved to Harvard in 1963 as an assistant professor. Obedience work continued, but the department did not grant him tenure, and he left in 1967. The ethics controversy is usually named as the cause, and it was certainly a factor; departmental politics and Harvard's own view of the work mattered too. The American Psychological Association had delayed his membership application in 1963–64 citing the ethics of the experiments — an unusually pointed signal from the body he was later honoured by.

CUNY, 1967–1984

From 1967 until his death he was at the City University of New York Graduate Center, where he built a doctoral programme in social psychology and did the second phase of his career. This is where the methodological inventiveness became the point rather than a means. He developed the lost letter technique — scattering stamped, addressed letters around a city and measuring how many are posted, as an unobtrusive measure of attitude toward the addressee. He ran the small-world experiments (1967), sending packages through chains of acquaintances and finding that any two people in the United States were separated by a median of about six steps — the origin of "six degrees of separation". He studied the familiar stranger (1972), the person you see daily and never speak to. He trained "cyranoids" (1977), people who speak lines fed to them by someone else, as a way of separating the source of a message from its content.

He published Obedience to Authority: An Experimental View in 1974. He died of a heart attack on 20 December 1984, aged fifty-one.

Major Contributions to Psychology

The obedience experiments

The finding is usually given as a number and should be given as a set of conditions. Milgram's central result was that obedience is not a constant but a function of situational structure, and that it can be moved anywhere between almost nothing and almost everything by changing features of the setting.

In the baseline condition — learner in another room, heard but not seen, experimenter present — 65% of participants went to 450 volts. In the variations:

Proximity. When the teacher and learner were in the same room, obedience fell to about 40%. When the teacher had to hold the learner's hand on the shock plate, it fell to about 30%.
Prestige of the setting. Moving the study from Yale to a run-down office in Bridgeport, Connecticut, lowered obedience, though not dramatically.
Dissenting peers. When two confederate teachers refused to continue, the great majority of real participants also refused — the strongest single effect in the whole series.
Conflicting authorities. When two experimenters disagreed about whether to continue, participants generally stopped.
Immediacy of the victim. The further removed the consequences, the easier the obedience.

Milgram also recorded the tension: participants sweated, trembled, stuttered, laughed nervously, dug their nails into their hands. The obedience was not calm. Yet about 84% of participants said afterward, in a follow-up questionnaire, that they were glad to have taken part.

The agentic state

Milgram's explanation was that under certain conditions a person shifts from an autonomous state, in which they take responsibility for their actions, into an agentic state, in which they experience themselves as the instrument of someone else's will and therefore do not regard themselves as the author of what they do. The shift is enabled by what he called binding factors — the gradual, incremental structure of the task, the sense of obligation to the experimenter, the difficulty of finding an exit — and the responsibility is displaced onto the authority, who is perceived as having assumed it.

The concept has been substantially displaced by later reinterpretation (see below), and it is fair to say it is now taught more as a historical position than as the current account.

The small-world method

The 1967 study is the one piece of Milgram's work whose influence has grown rather than shrunk. He asked people in Nebraska and Boston to get a letter to a specified target person in Massachusetts by forwarding it to someone they knew on a first-name basis who might be closer. Completed chains averaged around six links. The design was crude by later standards and the completion rate was low, but the demonstration that real social networks have short global path lengths became, through the mathematical work of Duncan Watts and Steven Strogatz in the 1990s, the foundation of network science — now used in epidemiology, information diffusion and the analysis of online platforms.

Methodological invention

Beyond the specific findings, Milgram's legacy includes a set of techniques for studying behaviour people would not report honestly: the staged situation, the confederate, the unobtrusive measure, the lost letter, the familiar stranger. Along with Asch's conformity research and Latané and Darley's bystander work, it established the situationist position in social psychology — that behaviour is a function of the situation far more than most people, including most psychologists, assume.

Major Publications

"Behavioral Study of Obedience" (1963, Journal of Abnormal and Social Psychology) — the baseline experiment
Obedience to Authority: An Experimental View (1974) — the full account of the series and the theory
"The Small-World Problem" (1967, Psychology Today) — the six-degrees result
The Individual in a Social World: Essays and Experiments (1977) — the collected papers
Television and Antisocial Behavior (1973, with R. Lance Shotland) — field experiments on media effects

Impact and Legacy

Milgram did two things at once, and they pull in opposite directions.

He produced the most cited demonstration in social psychology, and it is now standard content in medical ethics training, military leadership courses and business ethics curricula — used not as proof that obedience is automatic but as a warning that authority structures change the moral arithmetic of ordinary people. The Nuremberg defence is still mounted in war crimes trials; the psychology of harmful compliance in hierarchies — junior nurses overriding their judgment because a senior doctor gave the instruction — is his subject. And the obedience studies, with Zimbardo's prison study and the bystander work, anchored the situationist position that dominated social psychology for two decades.

He also caused the ethical apparatus that now governs human research. Diana Baumrind's 1964 critique in American Psychologist forced the discipline to state what it owed its participants, and the framework of informed consent, debriefing and institutional review that operates worldwide is substantially a response to his studies. A fair summary is that the field gained an empirical demonstration and the machinery ensuring the demonstration could not be repeated in its original form.

Criticism

The ethics. The objections were immediate and remain valid. Participants were deceived about the nature of the study; they were not told they could leave without penalty — and Milgram's own prods explicitly removed that option in the last instance; they were placed in a situation of acute distress, and several had to be told, mid-procedure, that no one was being hurt. Baumrind's 1964 argument was that the study treated participants as means, that the deception was unusually elaborate, and that the distress was not incidental but the mechanism. Milgram replied that the participants had consented to being in an experiment, that their distress resolved, and that follow-up showed the great majority were glad to have taken part. That reply does not meet the objection, and the study could not be run today.

The 65% is one condition, not the finding. Across more than twenty variations, obedience ranged from 0% to 92%. The headline number is the baseline condition and it is the one that got quoted, in Milgram's own writing and everywhere since.

Many participants did not believe it. This is the most damaging of the archival findings and the one that undercuts the inference most directly. Gina Perry, working through Milgram's papers and the Yale audiotapes, concluded that only about half of the participants fully believed the shocks were real — and of those who did, 66% disobeyed. If a substantial fraction of participants were playing along with what they took to be a theatre exercise, then the experiment does not demonstrate a willingness to harm a stranger under orders; it demonstrates something closer to a willingness to continue a performance. Some participants said so at the debriefing.

The procedure was not standardised. Perry's archival audio shows experimenters departing frequently from the four-prod script — improvising additional pressure, reassurance and threats, sometimes a dozen or more interventions in a single session. The study's internal validity depends on the prods being constant, and they were not.

An unpublished condition inverted the result. Among the experiments Milgram ran in 1962 was a condition in which the teacher and learner were relatives or close acquaintances. When the learner protested, roughly 85% refused to continue — nearly the inverse of the headline finding. It was never published.

Debriefing was inadequate. Some participants left convinced they had actually hurt someone; some did not learn the truth until months or years later, in some cases by posted letter. Perry's account does not overturn Milgram's claims but shows the published version was considerably tidier than the laboratory.

And the explanation has been replaced. Stephen Reicher and Alexander Haslam, working from Milgram's own data and audiotapes, argued for an engaged followership reading: participants continued not because they had switched off their agency but because they identified with the experimenter's project and had been recruited into it as collaborators. On that reading, the effective prod is not "you must" but "the experiment requires that you continue" — an appeal to shared purpose, which is why it worked better than direct orders, and why participants who identified most strongly with the scientific mission obeyed most. The agentic state has become correspondingly less central in current teaching.

The engaged followership reading has finally been tested directly, and it has not gone all one way. Reicher, Haslam and Birney restated the case in 2024 as a full research programme (Philosophia Scientiae 28(2):91–105), and the two years after produced adversaries. Grzyb and Dolinski (2025, Frontiers in Psychology 16:1613507) put 80 participants through an "obedience lite" procedure in which the participant either merely examined the learner or actually taught him for eight minutes first; 90% of examiners and 77.5% of teachers continued to 150 volts, and the teachers needed significantly more urging to keep going. The paper also collects the cases the model handles badly: in Milgram's most-cited new baseline condition experts rated identification with the learner as higher, yet obedience was among the highest in the series at 62.5%; one of the studies most often cited in support of the account (Haslam et al., 2014) involved no moral dilemma, no harm and no genuine authority pressure because it was run online; and when Grzyb and colleagues reframed the same task as commercial marketing research rather than a contribution to science, obedience was just as high. Grzyb, Dolinski and Cantarero (2025, Social Psychology 56(2):85–97) further found that the experimenter's gender made no difference — 88% compliance with a female experimenter against 90% with a male one in the laboratory, and no effect in a pre-registered scenario study with 793 participants either. The result is a live argument rather than a settled replacement.

And the agentic state has been given a neural test, with mixed results. In a paradigm in which people were ordered to administer pain to someone else, Emilie Caspar, Patrick Haggard and colleagues found that coercion shrank the temporal binding that normally marks a voluntary action — obliged participants experienced the action as less their own (Caspar, Christensen, Cleeremans & Haggard, 2016, Current Biology 26:585–592), which is close to what Milgram proposed. Their follow-up with functional imaging did not reproduce that loss at the group level inside the scanner (free choice 72.6 ms against coercion 74.0 ms in 25 participants), though stated responsibility still dropped sharply under coercion; what predicted who kept a sense of authorship was greater activity in the medial frontal gyrus, which tracked both preserved agency and preserved responsibility (Caspar, Beyer, Cleeremans & Haggard, 2021, PLOS ONE 16(10):e0258884). Read together the two experiments qualify rather than rescue Milgram's explanation: being ordered to hurt someone can dissolve the experience of being the author of the act, but it does not do so automatically, and the difference between those who lose it and those who keep it is measurable.

The generalisation to history is contested. Milgram's sample was an all-male volunteer pool from one American city, the procedure lasted an hour, and the participants knew they were in a Yale laboratory. Extrapolating from it to the Holocaust — which took years, involved career structures, ideology, propaganda and territorial war — is a very large inference, and Milgram himself was more cautious about it in the book than his framing elsewhere suggested.

And the phenomenon does appear to replicate. Jerry Burger's 2009 partial replication at Santa Clara University, designed to meet modern ethical constraints by stopping at 150 volts — the point at which the learner first demands to be released — found that 70% of participants continued past that point, statistically indistinguishable from Milgram's 82.5% at the same decision point. A Polish replication (Doliński et al., 2017) found higher rates still. The conclusion on the current evidence is that the effect is real, that it has not diminished over fifty years, and that the number people memorise is not the right one.

Associated Figures

Solomon Asch — whose conformity studies were the model, and whose research assistant Milgram was
Gordon Allport — his doctoral supervisor at Harvard
Philip Zimbardo — whose prison study is the companion piece, and suffered a comparable archival reckoning
Bibb Latané and John Darley — whose bystander work extended the situationist programme
Diana Baumrind — whose 1964 critique in American Psychologist is the foundational statement of what the study got wrong ethically
Herbert Kelman — among the earliest critics of the ethics of deception in research
Gina Perry, Stephen Reicher and Alexander Haslam — the archival and theoretical re-readings that revised the standard account

Learn More

References

Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4), 371-378.
Milgram, S. (1974). Obedience to Authority. Harper & Row.