Milgram's obedience experiment, and what the 65% actually means

In the baseline condition of Milgram's obedience experiment, 26 of the 40 participants — 65% — administered what they believed was the maximum 450-volt shock to another person. The shocks were fake; the instruction was not. An experimenter in a grey lab coat told them to keep going, and most of them did. Before the study began, psychiatrists had predicted that about 0.1% of people would go that far.

That gap is why the study is still taught, and also why it is so often misreported. The 65% came from one condition, not from human nature: across Milgram's variations the rate moved between 0% and 92.5%. Later archival research showed that parts of the standardised procedure were less standardised than the papers suggested.

• Milgram ran the baseline study at Yale University in 1961–62 with 40 men recruited through newspaper advertisements, who were told it was a study of memory and learning.
• The draw that assigned the roles of teacher and learner was rigged: the participant was always the teacher, and the learner was an actor, Mr. Wallace.
• The shock generator was fake, but a real 45-volt sample to the wrist made it feel genuine.
• 26 of the 40 participants (65%) went to 450 volts; all 40 reached at least 300 volts, and 14 stopped before the end.
• Obedience fell as the learner became more present (30% in the touch condition), and when the experimenter gave orders by telephone (20.5%) or two peers refused to continue (10%). It rose to 92.5% when someone else pressed the switches.

What did Milgram's obedience experiment test?

The study tested obedience: how far a person would follow an authority figure's instructions when those instructions conflicted with their own judgement. In Milgram's setup, following orders meant continuing to shock another person, one 15-volt step at a time, while that person protested.

Milgram originally set out to compare obedience in the United States with obedience in Germany, suspecting that something in German culture might help explain Nazi atrocities. He tested Americans first as a low-obedience baseline, and the results came back so high that he abandoned the comparison.

How was Milgram's experiment conducted?

The procedure matters, because most of the misunderstandings live in the details.

Recruitment and the rigged draw

Milgram recruited 40 men aged 20 to 50 from New Haven and the surrounding communities through newspaper advertisements and direct mail, from a wide range of occupations and education levels. Each was paid $4.50 for attending, and told the money was for turning up, whatever happened after that.

The cover story was a study of memory and learning at Yale. At the laboratory the participant met another volunteer — in fact a confederate, Mr. Wallace, a 47-year-old Irish-American accountant — and the experimenter, a stern school biology teacher in a grey technician's coat.

Roles were assigned by drawing slips from a hat, but the draw was rigged: both slips said "Teacher", and the participant always drew first, so the participant was always the teacher and the confederate always the learner. The learner was strapped into a chair in an adjacent room with electrodes on his wrist, while the participant sat down in front of the shock generator.

The fake generator and the learner's script

The generator had 30 switches, one for every 15 volts from 15 to 450, labelled from "Slight Shock" to "Danger: Severe Shock", with the last two marked "XXX". It produced no shocks at all. A real 45-volt sample to the wrist made it feel genuine.

At the machine, the teacher read word pairs and the learner answered with one of four switches. Every wrong answer earned a shock one step higher, and the learner's answers were scripted to give about three wrong answers for every correct one. His distress was scripted too:

• 75 volts: the first audible grunts of pain.
• 120 volts: louder, more insistent protests.
• 150 volts: a complaint about a heart condition.
• 300 volts: pounding on the wall of the neighbouring room.
• After about 330 volts: silence.

The experimenter's rules covered the rest: wait 5 to 10 seconds before treating silence as wrong, announce the voltage, and raise the level one step for every mistake.

The four prods

When a participant hesitated or refused, the experimenter had a published script of four prods, used in order, with the sequence restarted each time the participant pushed back:

1. 1."Please continue."
2. 2."The experiment requires that you continue."
3. 3."It is absolutely essential that you continue."
4. 4."You have no other choice, you must go on."

The tone was firm but not impolite, and the script was presented as standardised. On that point, the archive and the published account do not fully agree.

What were the results?

26 of the 40 men went all the way to 450 volts. All 40 continued to at least 300 volts, and the 14 who stopped did so at different points between 300 and 375 volts. Psychiatrists surveyed beforehand had predicted 0.1%.

The number hides what the sessions were like. No participant sat through it comfortably: all questioned the procedure at some point, many pleaded to stop or offered to return the money, and the recorded signs of tension included sweating, trembling, stuttering and nervous laughter. During debriefing, participants rated the final shocks on a 14-point pain scale; the most common answer was 14, "extremely painful". They believed it.

Which variations mattered?

Between 1961 and 1962 Milgram ran at least 18 variations of the study, with 636 participants in total. Each changed one feature of the situation.

Read together, the variations describe obedience as a function of the situation rather than a trait: it fell as the learner became more present, as the authority became more distant or less coherent, and as other people visibly refused; it rose when the participant could hand the pressing itself to someone else. That last effect is what Milgram called diffusion of responsibility — when no single person feels fully responsible, compliance gets easier.

What the 65% does not mean

Three readings get the number wrong.

It does not mean that two-thirds of people will torture someone on command. The 65% describes what 40 men did in one laboratory situation; 35% refused even there, and the variations show how much the rate depends on the setup. A finding that swings from 0% to 92.5% with changes to the room is not a fixed human constant.

It does not mean the participants were cruel, or that anyone was hurt. No shocks reached the learner, who was an actor, and the generator was a prop. What was real was the participants' belief that they were hurting someone — and their distress is part of the record, not an afterthought.

It does not mean that obedience and conformity are interchangeable terms. Conformity is adjusting to what a group does or believes; obedience is doing what an authority tells you. Milgram's participants were not copying a crowd — each sat alone with an experimenter whose role was to keep them going.

The real-world parallel, and its limits

Milgram himself drew the parallel to the Holocaust, asking whether Eichmann and his accomplices could be explained away as men who were just following orders. Historians and psychologists have since argued that the parallel only goes so far: in the laboratory, participants were told the shocks would cause no lasting harm, the encounter lasted less than an hour, and the authority was present throughout, while genocidal violence involves ideology, propaganda, prolonged brutality and severe consequences for anyone who refuses to take part. The experiment shows one mechanism at work; it does not explain the whole of atrocity.

Why did so many people obey?

Most participants did not obey cheerfully. In interviews afterwards, people described themselves as tormented by what they believed they were doing, and many justified continuing by saying they were "doing their job" or "helping science" — the after-the-fact reasoning familiar from cognitive dissonance, in which behaviour that has already happened pulls the belief into line behind it.

Agency theory

Milgram described two states a person can be in. In the autonomous state, you direct your own actions and carry responsibility for them; in the agentic state, you act as an instrument of someone else's will, and responsibility shifts to them. Two things must be in place for the shift, in his account: the person giving the orders must look qualified to give them, and you must believe they will take responsibility for the outcome. His evidence fit: when participants were reminded that the responsibility was their own, almost none went on; when the experimenter said that he would take it, many who had been refusing continued.

The cues that make an authority look legitimate are quick and often invisible: a grey lab coat, a university name, the assumption that someone else is in charge of safety. Milgram's variations tested them directly — obedience fell by nearly twenty points when the study moved off campus. These are the shortcuts described in our overview of cognitive biases: efficient, usually useful, and capable of running without your noticing.

Engaged followership: identification rather than blind orders

The popular version of the result is a story about blind submission: ordinary people, faced with a confident man in a uniform, lose the ability to say no. Archival analyses by S. Alexander Haslam and Stephen Reicher support a different reading. On their engaged followership account, participants did not simply cave; they identified with the experimenter's scientific mission and kept going because they believed the research mattered. The prods are the evidence: appeals to the value of the work — "The experiment requires that you continue" — worked better than the flat command "You have no other choice, you must go on", which participants frequently resisted.

The prods in the paper were not the prods on the tapes

Milgram presented the study as standardised: the same script, the same four prods, delivered the same way. The archived recordings tell a messier story.

Gina Perry's 2013 archival analysis of the audiotapes found that experimenters often improvised when participants resisted, going beyond the published script with prompts more coercive than anything in it. The improvisation varied between sessions and participants, and was especially common in the sessions with women — a series in which the study found no gender difference in obedience, suggesting the extra pressure influenced the results. Stephen Gibson's analysis of the recordings found a similar drift: prods were combined and reworded rather than read out verbatim, so the classic list worked more like a repertoire the experimenter drew on.

None of this makes the study a fabrication. The point is that the tidy textbook version of a uniform procedure is itself a simplification, and that scientific reporting can smooth over the messier reality of a laboratory. Most short summaries still present the prods as a fixed script; the archive is the correction.

Why the experiment is still called unethical

The study is described as unethical for what happened in the room and for when it happened. In 1961 the formal safeguards that now govern psychological research did not exist, which explains the study without settling the argument about it.

Consent, deception and the right to withdraw

Participants volunteered for a study of memory and learning. The shocks, the learner's pain and the real purpose of the study were all hidden from them, so the consent they gave could not cover what they were actually asked to do. Milgram defended the deception on the grounds that some truths cannot be reached without it, and that he disclosed it afterwards. The other fault line is withdrawal: the four prods were written to push a wavering participant to continue, and critics read them as pressure against the withdrawal the study claimed to allow. Milgram answered with the data — 14 of the 40 (35%) did stop, so refusal was possible. Both claims can be true: the option existed, and the script made it harder to take.

The debriefing dispute

Distress was not rare in the sessions: trembling, sweating, stuttering, nervous laughter, pleas to stop. Milgram argued that the effects were short-lived. He debriefed participants, followed up a year later, and reported that 83.7% were glad to have taken part, 74% said they had learned something of personal importance, and 1.3% wished they had not been involved.

The archival record complicates that defence. Perry found that many participants were not properly debriefed soon after the study, contrary to Milgram's published account, and that some learned only months later, by post, that the shocks had never been real. Her work also documents distress that outlasted the sessions. That is why the ethics of the study is still argued rather than closed.

What changed

The study's most durable legacy is procedural. Institutional review boards, informed consent requirements, limits on deception, an explicit right to withdraw at any time and mandatory debriefing all became standard in the decades that followed, with the Milgram controversy among the cases that pushed them into place. Both the American Psychological Association and the British Psychological Society now require protections the participants of 1961 did not have. It is taught as much through its problems as its findings: a study that changed psychology and the rules under which psychology works.

Terminology for the exam: obedience, the agentic state, diffusion of responsibility and the four prods blur together when you revise them one at a time, and they stick better when you learn them against each other — the approach in our guide to memorising psychology terms. To keep them ready, save "obedience", "agentic state" and "prods" to a Social psychology list in Psychology World and drill them.