← Concept Vault
Concept · Psychology · Maps to the Law of Cause and Effect

Obedience (Milgram)

Ordinary people, given a plausible authority and a slow ramp, will do extraordinary harm — but the real finding is messier, more conditional, and more interesting than the legend that ate it.
Read · 21 minPillar · Psychology

This is the most famous experiment in the history of social psychology, and also one of the most distorted. Nearly everyone "knows" Milgram: people in a lab kept flipping switches to shock a screaming stranger because a man in a coat told them to, and most went all the way to a voltage marked lethal. That story is roughly true and badly incomplete. The genuine finding is real, robust, and disturbing. The story built on top of it — that the experiment proved we're all latent Nazis one instruction away from atrocity — is an oversell that the data don't support and that Milgram himself half-built, half-fought. The work here is separating the steel core from the mythology welded onto it, because both the finding and its distortion teach you something about how the mind handles authority. Let's take it all the way down.

01

What it is (and where it came from)

Stanley Milgram ran his obedience studies at Yale University between 1961 and 1963, with the first paper — "Behavioral Study of Obedience" — published in the Journal of Abnormal and Social Psychology in 1963. He expanded the work in his 1974 book Obedience to Authority: An Experimental View. The timing is not incidental. The first sessions began in July 1961, three months after the trial of Adolf Eichmann opened in Jerusalem. Milgram, the son of Jewish immigrants, was openly chasing the question the trial had forced on the West: how does a continent's worth of seemingly normal people participate in the machinery of genocide? The conventional answer in 1961 was that Germans were somehow different — more authoritarian, more cruel, a defective national character. Milgram designed a machine to test whether the explanation was in the people or in the situation.

The setup, in its canonical form: a volunteer arrives at a Yale lab, recruited by newspaper ad for a study on "memory and learning," paid $4.50. He meets the experimenter (a stern man in a gray lab coat) and another "volunteer" (actually a confederate, a mild middle-aged accountant named Mr. Wallace). A rigged drawing always makes the real subject the "Teacher" and the confederate the "Learner." The Learner is strapped into a chair in an adjacent room with an electrode on his arm. The Teacher sits at an imposing shock generator with 30 switches labeled from 15 volts ("Slight Shock") to 450 volts, the top switches ominously marked "Danger: Severe Shock" and finally just "XXX." Each time the Learner gets a word-pair wrong, the Teacher is to deliver a shock, climbing one level — 15 volts higher — with every error.

No one is actually shocked. The Learner's responses are scripted and pre-recorded. As the voltage climbs he grunts, then complains, then at 150 volts demands to be let out, then screams about his heart condition, then — past 300 volts — goes ominously silent, which the Teacher is told to treat as a wrong answer and shock anyway. When the Teacher hesitates, the experimenter delivers a graded series of scripted "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."

That's the apparatus. What Milgram wanted to know was a single number: how far up the board would ordinary people go.

02

The real science / the evidence

Here is the result that detonated. In the baseline condition — Learner in an adjacent room, audible but unseen — roughly 65% of subjects (26 of 40) went all the way to 450 volts, the maximum, past the screams and past the silence. Every single subject went to at least 300 volts before any of them refused. Milgram had polled psychiatrists beforehand; they predicted that only a pathological fraction of a percent — something like 1 in 1,000, a "lunatic fringe" — would reach the top. The professionals were off by more than two orders of magnitude. That gap between what experts predicted and what people actually did is the deepest part of the finding: we systematically underestimate the power of the situation and overestimate the power of individual conscience.

But the 65% number is the part everyone memorizes and the part that misleads. Milgram did not run one experiment. He ran a program of roughly two dozen conditions (counts vary by how you slice them — 18 or 19 described in the 1974 book, roughly 23 to 24 in the archives), somewhere on the order of 700–800 subjects across the whole program, and the whole point of the program was that the obedience rate moved violently depending on the situation. The number was never the finding. The sensitivity of the number was the finding. Consider the spread:

  • Proximity matters enormously. When the Learner was in the same room (face-to-face), full obedience dropped to about 40%. When the Teacher had to physically force the Learner's hand onto a shock plate (the "touch-proximity" condition), it fell to roughly 30%. Distance from the consequence is permission.
  • Authority proximity matters too. When the experimenter gave orders by telephone rather than in person, full obedience collapsed to around 20–21% — and notably, some subjects lied, telling the experimenter they were raising the voltage while secretly administering the lowest shock. Authority you can't see is far easier to defy.
  • The setting matters. When Milgram moved the study off the Yale campus to a run-down commercial office in nearby Bridgeport under the name "Research Associates of Bridgeport," obedience dropped to roughly 48% — lower, but, strikingly, still close to half, suggesting the institutional prestige of Yale was a factor but not the whole story.
  • Peers matter, in both directions. When two confederate "co-teachers" defied the experimenter and walked out, full obedience cratered to around 10% — the single most powerful lever for reducing obedience in the whole program. The presence of anyone modeling refusal gave subjects a script for rebellion. Conversely, when the subject only read the answers while a confederate flipped the final switch, obedience to 450V rose to over 90%: distance from the physical act, again.
  • Conflicting authority neutralizes. When two experimenters gave contradictory orders, obedience essentially collapsed — subjects stopped at the first sign of disagreement at the top. Authority works by appearing unified and legitimate; split it and the spell breaks.

So the honest version of "the finding" is not 65% of people will kill on command. It is: obedience to authority is shockingly high under the right conditions and shockingly fragile under others, and the variables that move it are situational — proximity, legitimacy, surveillance, and the presence of allies — far more than they are dispositional. The conditions that produced 65% were specifically engineered: a prestigious institution, a visibly legitimate scientist, a victim out of sight, a gradual ramp, and no model of resistance.

One more piece of evidence that's usually omitted: subjects did not obey calmly. Milgram's own films show men sweating, trembling, stuttering, laughing in nervous fits, begging the experimenter to check on the Learner. They were in acute distress. They obeyed while suffering over obeying. This is critical, because it kills the lazy reading that obedient subjects were cold or sadistic. They hated what they were doing and did it anyway. The horror is not that people enjoy hurting others on command — it's that they'll do it against their own visible, agonized conscience.

How robust is all this? More robust than most of 1960s social psychology, which is faint praise in a field gutted by the replication crisis. The crucial point: Milgram's core effect has actually been re-run, which most famous old studies have not. Jerry Burger replicated a modified version in 2009 (published in American Psychologist), stopping at 150 volts for ethical reasons — the point where the Learner first demands release, which Milgram's data showed was the key "point of no return" (most who passed 150V went all the way). Burger found obedience rates statistically indistinguishable from Milgram's, in a modern American sample, with full ethics-board oversight. Partial and variant replications and analogues across several countries have generally found high obedience too, though rates vary. So the basic effect — high compliance with a legitimate authority's escalating destructive orders — is one of the more replicable results in social psychology. That's the steel core. Now the rust.

03

The mechanism or model

Milgram's signature explanatory concept is the agentic state. His claim: human beings have two modes of operating. In the autonomous state, you act on your own goals and feel responsible for your conduct. In the agentic state, you reconceive yourself as an instrument carrying out another person's wishes, and you no longer feel responsible for the action — you feel responsible only for how well you execute the orders. The responsibility, in your own internal accounting, gets transferred upward to the authority. The recurring line from subjects, when asked who was responsible if the Learner was harmed, was a version of "He told me to. I was just doing what I was supposed to."

Milgram argued this is a deep, partly evolved feature: any animal that lives in hierarchies needs a mechanism for subordinating individual judgment to the group's command structure, because coordinated hierarchies out-compete crowds of autonomous individuals. The agentic shift is the psychological switch that makes hierarchy possible. The cost is that it can be triggered by a legitimate-seeming authority and pointed at destructive ends. On this model, the subject isn't a monster and isn't even fully a free agent in the moment — he has entered a different mode of self-understanding in which the moral weight of the act is, from the inside, no longer his.

Supporting machinery that Milgram and later theorists identified, much of which is sturdier than the agentic-state concept itself:

  • Gradualism / the foot-in-the-door ramp. No one is asked to deliver 450 volts cold. They're asked for 15, having already done 0; then 30, having already done 15. Each step is trivially larger than the last, and each step is made consistent by the one before it — to refuse at 195 volts is to implicitly admit that 180 was already wrong. The escalation manufactures its own justification. This is commitment and consistency doing brutal work, and it's one of the best-supported principles in all of persuasion research.
  • Diffusion and transfer of responsibility. The authority explicitly absorbs the blame ("I'm responsible, not you"), which licenses the action. Bystander research (Latané and Darley, late 1960s) independently confirmed how readily humans shed felt responsibility when someone else appears to hold it.
  • The legitimacy of the frame. Subjects had voluntarily entered a scientific context at Yale with a stated benevolent purpose. They trusted that a real institution would not actually let someone be killed for a memory study. Much of the obedience may be less "I will hurt this man" and more "this credentialed system surely knows what it's doing and wouldn't permit real harm." That's not blind submission — it's a (badly misplaced) inference about institutional safety.
  • No exit script. The subjects were not given language for refusing. They didn't know how to quit — there was no form, no script, no model. This is exactly why the "rebel peers" condition was so devastating to obedience: it supplied the missing template.

A strong modern reinterpretation — "engaged followership," advanced by Stephen Reicher and Alex Haslam in the 2010s — directly challenges the agentic-state story. Reicher and Haslam (building on their broader critique of situationism, including their 2002 BBC Prison Study) re-analyzed Milgram's own recordings and prods and argued that obedient subjects were not passive automatons who'd surrendered their will. The prods themselves are revealing: the final, most commanding prod — "You have no other choice, you must go on" — was the least effective. When the experimenter framed continuing as an order, subjects tended to refuse. They were most likely to continue when persuaded that doing so served a worthy collective project — science, knowledge, the experiment's value. On this view, the obedient subject is an active, identified follower who has bought into the authority's mission and chosen to advance it, not a will-less agent. The mechanism isn't surrender of self; it's recruitment of self into a cause framed as good. That is a meaningfully different — and arguably more frightening — story, because it means atrocity runs on belief and identification, not just submission.

04

Sharp examples

The reason Milgram still grips is that the mechanism is everywhere once you can see it, and it almost never looks like a lab.

The Eichmann frame that started it. Hannah Arendt covered the Eichmann trial for The New Yorker and coined "the banality of evil" in 1963 — her thesis that Eichmann was not a frothing ideological monster but a careerist bureaucrat who organized deportations the way one might optimize a shipping schedule, never quite confronting what he was doing. Milgram's lab and Arendt's courtroom became intellectual twins: both arguing that mass atrocity does not require monsters, only ordinary people inside a system. (Honest caveat: later historical work, especially *Bettina Stangneth's Eichmann Before Jerusalem, 2011, used Eichmann's own private writings to argue he was a far more committed, ideologically motivated antisemite than the courtroom performance suggested. The "banality" thesis is contested as biography even as it survives as a description of how systems recruit ordinary participants.* Hold both.)

The destructive-obedience disasters that map cleanly onto the variables. Consider aviation's "captain authority" crashes — cases where a co-pilot watched a captain fly into a fatal error and did not forcefully intervene, deferring to rank up to the moment of impact. The entire discipline of Crew Resource Management was built in response, explicitly to give subordinates a script and a duty to challenge authority — which is the Milgram "rebel peer / break the spell" lever, institutionalized. The medical analogue: the "hierarchy gradient" in operating rooms, where nurses historically would not contradict a surgeon's obvious error. The fix, again, is procedural permission to defy — checklists, mandatory call-outs, "stop the line" authority.

The con-artist and compliance versions. The "strip-search phone scam" of the 1990s and early 2000s — in which a caller impersonating a police officer phoned fast-food restaurants and persuaded managers to strip-search and abuse employees over the phone, purely on claimed authority — is Milgram with no lab coat and no Yale, just a voice asserting legitimacy plus gradual escalation. It was attempted more than 70 times across the US, and often worked. Telephone authority, in Milgram's own data, was the weak condition — and it still worked here, because the other levers (gradualism, an authoritative frame, no model of refusal, a victim whose suffering the manager could rationalize as "official") were all present.

The everyday version, which is the point. You do not need genocide. You need only to notice the moment you do something you mildly disagree with because someone with a title or a clipboard told you to, and you feel the responsibility quietly slide off your shoulders onto theirs. The agentic shift is the small daily anesthetic that lets organizations function — and occasionally lets them commit horrors. The same switch that makes you a functional employee makes you a potential instrument.

05

Where it gets distorted or oversold (the honest cut)

This is the most important section, because Milgram is overcited and undercriticized in popular culture, and the serious critiques are devastating to the legend without erasing the core.

*Gina Perry's Behind the Shock Machine (2012)* is the essential demolition of the mythology. Perry, a psychologist, went into Milgram's own archives at Yale (the boxes of tapes, notes, and subject correspondence) and surfaced findings that complicate the clean textbook story:

  • The script was not fixed. Milgram and his experimenter (a man named John Williams, a high-school biology teacher, not a psychologist) frequently went off-script — improvising extra prods, pressuring, cajoling, sometimes well beyond the four standardized lines. The "standardized" four prods that appear in every textbook were not always what subjects actually heard. If the experimenter ad-libbed coercion, the experiment was less a clean measure of obedience-to-standard-authority and more a measure of how much a determined live human could push another.
  • Many subjects didn't believe it was real. Perry found, in Milgram's own post-experiment data, that a substantial fraction of participants reported they suspected or were sure the shocks were fake — that the man wasn't really being hurt. Milgram had this information and tended to downplay it in his published accounts. If a large share of "obedient" subjects went to 450V partly because they sensed it was theater, the demonstration of genuine destructive obedience is significantly weaker than advertised. (This is contested — other analyses argue most subjects did believe in the moment, and the stress on the films is hard to fake — but the doubt is real and was suppressed, which is itself damning.)
  • The debriefing was often inadequate or absent for a long time. Many subjects left believing they might have genuinely hurt or killed a man, and were not promptly told otherwise. Milgram's claim that participants were carefully dehoaxed and even grateful is, per the archives, partly retrospective gloss.
  • Selective reporting. Milgram foregrounded the conditions and framings that produced the most striking obedience and a clean narrative; the program was messier, the variations more numerous and contradictory, than the iconic "65%" presentation implies.

Then there's the ethics, which is not a footnote — it's a turning point in the history of science. Milgram deceived subjects, exposed them to extreme acute psychological distress (some convinced they'd killed someone), used escalating coercion to keep them in their seats, and arguably did not give them a real ability to withdraw despite later claiming they could leave anytime — the entire experiment depended on overriding their attempts to leave. The blowback (a sharp 1964 critique by Diana Baumrind in American Psychologist led the charge) helped drive the creation of modern Institutional Review Boards and informed-consent norms. A consequence worth sitting with: the study could not be run today in its original form, which means the famous 65% figure is frozen in a methodology we now consider unethical and cannot cleanly redo. (Burger's 2009 partial replication exists precisely because the full thing is off-limits.)

What the experiment does and does not show. It does show: under specific, engineered conditions of legitimate authority, gradual escalation, victim distance, and absence of dissenting models, a large fraction of ordinary people will continue inflicting what they believe to be serious harm, against their own conscience and visible distress. That is real and replicated. It does not show: that "everyone is 65% likely to commit atrocity," that obedience is a fixed human constant, that the people who obeyed were morally defective, or that situation fully erases disposition. It does not show that the agentic-state model is correct (that specific mechanism is contested and partly supplanted by the followership account). And it does not show what the meme implies — that the Holocaust was "just obedience." Genocide involved ideology, dehumanization, propaganda, opportunism, and active belief; reducing it to switch-flipping is a category error Milgram's data don't license even if his rhetoric sometimes leaned that way.

The deepest distortion is dispositional smuggling in reverse. People wield Milgram to feel enlightened — "I now know I'd resist, unlike those sheep" — which is exactly the fundamental attribution error the study should cure. The lesson isn't "look how weak those people were." It's "the people who obeyed were demographically and psychologically you, and the only reliable protections are situational: visible exits, dissenting allies, split or de-legitimized authority, and proximity to consequences."

06

The debates & frontier

The live questions, marked by how settled they are:

  • Agentic state vs. engaged followership (genuinely open). Did obedient subjects surrender their will to an authority, or actively identify with a cause they were persuaded was good? Reicher and Haslam's re-analysis of the prod data — that the commanding "you have no choice" prod failed while the cause-invoking "the experiment requires you continue" worked — is strong circumstantial evidence for followership over passive submission. This is not resolved; it's one of the more interesting open arguments in social psychology, and it changes the practical lesson (resist by de-legitimizing the cause, not just by "saying no to authority").
  • How much did belief matter? (contested, important). Perry's archival finding that many subjects doubted the shocks' reality cuts directly at construct validity. The counter-argument — the filmed distress, the physiological stress, and the fact that even suspecting it's fake doesn't fully explain the agony — keeps the effect alive but wounded. An honest student holds the effect as real but with a credibility-of-the-victim confound that was underreported.
  • Replication and generalizability (mostly favorable, with caveats). Burger 2009 is the gold-standard modern replication and it held. But it stopped at 150V, it screened out vulnerable participants, and "150V predicts 450V" is an inference, not a direct re-demonstration of the full lethal ceiling. Cross-cultural variation is real but under-studied with rigorous designs. Verdict: the direction and rough magnitude of the effect are robust; the precise "65%" is condition-specific and should never be cited as a universal human rate.
  • Ecological validity (a permanent asterisk). A one-hour lab with a fake machine and a benevolent scientific frame is not a regime, a war, or a workplace. The mapping from lab obedience to real atrocity is suggestive, not proven. Milgram built a powerful demonstration; he did not build a complete theory of how nations commit mass murder, and the field has largely stopped pretending he did.
  • The frontier has moved toward (a) immersive/virtual replications — running Milgram-style paradigms with virtual or clearly-known-to-be-fake victims, where subjects still show stress and partial obedience, which interestingly suggests something beyond simple belief-in-harm is operating; and (b) integrating the obedience literature with the followership, social-identity, and institutional-design literatures to ask the actionable question: not "why do people obey?" but "what specific structures make defiance available and likely?" That's where the science is genuinely useful now.
07

How to apply it

The point of understanding this is not trivia — it's building yourself and your environments to be defiance-capable when it counts.

  1. Distrust the agentic slide in yourself. The diagnostic sensation is relief — the feeling of responsibility sliding off you onto an authority. "I'm just doing my job / following orders / doing what the system requires." That exact felt-relief is the alarm. The moment you stop owning the outcome and start owning only your compliance, you've entered the dangerous mode. Name it when you feel it.
  1. Re-attach yourself to the consequence. Obedience to harm runs on distance. The single most reliable internal counter is to mentally collapse the proximity: picture the actual human on the other end of the action, in the same room, in front of you. Milgram's own data prove this works — proximity to the victim roughly halved obedience. Force yourself to face the consequence and the agentic spell weakens.
  1. Engineer dissenting allies — they are the strongest lever. The "rebel peers" condition dropped obedience to ~10%. One other person modeling refusal gives everyone a script. In any system you build or join, the protection against destructive groupthink is structurally guaranteed dissent: someone whose explicit job is to object, a culture where the junior person can stop the line. This is why Crew Resource Management and surgical checklists exist. If you run a team, your job is to make refusal normal and cheap, not heroic and expensive.
  1. Interrogate legitimacy and unity of the authority. Obedience collapsed when authorities conflicted or were de-legitimized (off-campus, by phone). Practically: ask who actually says so, and do all the relevant authorities agree? Manufactured authority (the fake cop on the phone, the confident con) relies on you not checking. Splitting or verifying authority breaks the trance.
  1. Refuse the first small step, not the big one. The ramp is the trap. By the time you're at the morally obvious refusal point, you've already built a wall of self-justifying prior steps. The cheap place to defy is early, at the first request that pings your conscience, before consistency-pressure locks in. Treat the small "harmless" compliance as the real decision point.
  1. Stop using it to feel superior. The only honest application starts with "I am one of the 65%, under the right conditions." Humility about your own situational vulnerability is the precondition for actually building the guardrails. The person certain they'd resist is the least protected.
08

How It Connects

The Law of Cause and Effect (primary map). This is the cleanest link, and it cuts two ways. First, Milgram is a brutal demonstration of the law operating through a human who has talked himself out of being a cause. The agentic state is, precisely, a denial of one's own causal authorship — "I didn't cause his suffering, the experiment did, I was only a link." But Cause and Effect is non-negotiable: the man flipping the switch is a cause of the effect, regardless of how he narrates his own role. The experiment exposes the gap between the physics of causation (you acted, harm followed, you are a cause) and the psychology of felt responsibility (you can almost completely sever your sense of authorship while remaining the literal cause). The law says nothing escapes the chain; Milgram shows the mind's favorite escape hatch is to reassign its link in the chain to someone above it. Second, the experiment itself is a study in which causes actually move the effect — Milgram's whole program was running the same outcome (a person's obedience) against different antecedent conditions to isolate the real causal levers (proximity, legitimacy, peers) from the imagined ones (individual character). That is Cause and Effect used as a method: vary the cause, watch the effect, find what actually drives it. The practical synthesis: you remain a cause of everything your hands do, especially when an authority offers to relieve you of that fact — and the way to control the effect is to control the situational causes (build in exits, allies, proximity) before the moment arrives.

Related concepts & structures: Conformity (Solomon Asch's line-judgment studies, 1951 — obedience's quieter sibling, where the pressure is peer consensus rather than authority) · the Stanford Prison Experiment (Zimbardo, 1971 — the situationist cousin, and itself even more heavily critiqued for fabrication and coaching, so cite it more cautiously than Milgram, not less) · the Fundamental Attribution Error (we explain others' obedience by their character and our own by the situation — the exact bias Milgram should cure) · diffusion of responsibility and the bystander effect (Latané & Darley) · commitment & consistency and foot-in-the-door (the gradualism mechanism) · social identity / followership theory (Reicher & Haslam) · the banality of evil (Arendt) and its biographical contesting (Stangneth).

Thinkers: Stanley Milgram (the experiment and the agentic-state model) · Diana Baumrind (the founding ethics critique) · Gina Perry (the archival demolition of the clean narrative) · Stephen Reicher & Alex Haslam (the followership reinterpretation) · Jerry Burger (the modern replication) · Hannah Arendt (the philosophical frame) · Solomon Asch (Milgram's mentor, whom he assisted at Princeton, and the conformity precursor).

The Compression

The horror of Milgram isn't that monsters exist — it's that ordinary, conscience-stricken people will hurt a screaming stranger when a legitimate-seeming authority owns the responsibility, the victim is out of sight, the harm escalates one harmless step at a time, and no one else in the room is saying no. Change any one of those conditions and most people stop. Defiance is not a virtue you have; it's a situation you build.

Your Practice
My Depth
My Notes
← Previous
Conformity (Asch)
Next →
The Bystander Effect