Everyone "knows" the Milgram study: ordinary people shocked a stranger to death because a man in a lab coat told them to. That headline is true enough to be useful and simple enough to be misleading. This page keeps the real result — the effect is genuine and it has been replicated — while adding the two things most summaries skip: the variations that show obedience was never automatic, and the decades of archival and theoretical work that changed how we read the whole thing. Read it and you'll understand Milgram better than the person who only remembers "65%."
Bodhi says
The trap here is treating "65% obeyed" as the finding. It's the baseline. Milgram ran roughly two dozen variations, and the obedience rate swung from near-total to near-zero depending on the situation. The whole point of the research program was that a single knob — how close the authority is, how close the victim is, whether anyone else refused — moves ordinary behavior enormously. Memorize the pattern, not the percentage.
The paradigm, in one paragraph
Participants answered a newspaper ad for a study on "punishment and learning" at Yale and were paid $4.50 (Milgram, 1963). Through a rigged drawing, the real participant always became the teacher; a mild, likeable confederate — Mr. Wallace — became the learner. The teacher read word pairs and, for each wrong answer, delivered a shock from a generator labeled 15 to 450 volts, in 15-volt steps, with ominous captions climbing to "Danger: Severe Shock" and finally "XXX." No real shocks were ever given. As the voltage rose the learner protested, complained of a heart condition, screamed, demanded to be let out, and eventually went silent. When the teacher hesitated, the experimenter delivered a scripted prod: "Please continue," then "The experiment requires that you continue," then "You have no other choice, you must go on." Before the study, psychiatrists and laypeople alike predicted almost no one would go all the way — estimates hovered around 1%. In the baseline condition, about 65% went to 450 volts, and roughly 80% continued past the learner's cries about his heart (Milgram, 1963, 1974).
The single most important idea in this topic
Obedience was situationally elastic. The 65% baseline is one point on a curve. Move the authority to the phone, move the victim into the room, or let the participant watch someone else refuse, and obedience falls off a cliff. If you leave this page believing obedience is a fixed human constant, you've learned the opposite of what Milgram found.
The variations are the real study
Milgram didn't run one experiment; he ran a program of them, and the variations are where the science lives. Each one isolates a different situational force. Watch obedience move.
- Proximity of the learner. In the baseline, the learner was in another room, heard but unseen. Bring him into the same room and full obedience drops to about 40%. Require the teacher to physically press the learner's hand onto a shock plate — the "touch" condition — and it falls to roughly 30%. The closer the suffering, the harder to inflict.
- Proximity of the authority. When the experimenter left the room and gave orders by telephone, full obedience collapsed to about 20%, and some participants quietly lied — telling the experimenter they were raising the voltage while secretly delivering the lowest shocks. Authority at a distance is weak authority.
- Prestige of the setting. Milgram moved the study from Yale to a rundown commercial office in nearby Bridgeport, stripping away the university's institutional aura. Obedience dropped (to roughly 48%) but did not vanish — the lab coat and the framing still carried weight even without the ivy.
- Defiant peers. This is the most striking result of all. When the real participant worked alongside two confederate "teachers" who refused to continue — one quitting at 150 volts, the other at 210 — full obedience fell to about 10%. One visible model of resistance dismantled the situation's grip almost completely.
- Conflicting authorities. When two experimenters began to disagree about whether to proceed, obedience essentially disappeared: participants stopped. Nothing the victim did ever produced that effect — but the moment authority's own definition of the situation became unclear, people broke role (Milgram, 1974).
Notice the through-line. The learner's suffering, by itself, never stopped everyone. What reliably freed people was any crack in the authority structure — distance, a dissenting peer, an argument among the experimenters. That's the finding, and it's a more hopeful one than "65% would kill a stranger."
Why did they obey? Milgram's answer, and a better one
Milgram's own explanation was the agentic state: he argued that under legitimate authority people shift out of an autonomous mode, where they own their actions, into an "agentic" mode, where they see themselves as instruments carrying out someone else's will. Responsibility gets handed upward. It's a vivid idea, and it captures something real — many participants insisted afterward that the experimenter, not they, was accountable. But as a mechanism it has aged poorly. It treats obedience as a near-automatic mental switch, which fits neither the visible agony of the participants (people in an "autopilot" state don't sweat, tremble, and beg to stop) nor the variation data (an automatic switch shouldn't be flipped off by a colleague quitting the room).
There's a more defensible account: obedience was engineered by several ordinary, well-documented processes stacked together. The shocks rose in tiny 15-volt increments, so there was never a bright line to refuse at — a foot-in-the-door slide where each step is barely worse than the last. Responsibility was diffuse and, when questioned, explicitly assigned to the experimenter. The situation was novel and ambiguous, and participants leaned on the nearest expert exactly as informational influence predicts. And the "brisk pace" the experimenter set left no time to stop and think. None of that requires a mystical trance — it's the same influence machinery you meet everywhere else in social psychology, aimed all at once.
Where the field is now — the reappraisal
Milgram's work has been through a serious, healthy re-examination, and it's worth taking seriously rather than mythologizing. Gina Perry's archival research (Behind the Shock Machine) went back to Yale's records and interviewed surviving participants: she documented that the procedure varied more than the published reports suggest, that the experimenter often improvised beyond the four scripted prods, that a meaningful number of participants doubted the shocks were real, and that Milgram's debriefing was incomplete — some people left believing they had genuinely hurt a man (Perry, 2013). Separately, Haslam and Reicher have reinterpreted the results as engaged followership: people obeyed not on blind autopilot but because they came to identify with the scientific mission, and were most compliant when the experimenter appealed to the value of the research rather than simply barking orders (Haslam & Reicher, 2017). Crucially, none of this means the effect is fake — Jerry Burger's careful partial replication took participants only to 150 volts and still found about 70% ready to continue, statistically indistinguishable from Milgram's rate (Burger, 2009). The honest summary: the phenomenon is real, but "blind obedience to authority" is the wrong caption. This is science maturing. For how the field handles cases like this more broadly, see Science Corrects Itself.
Blind agentic-state reading
Legitimate authority flips a mental switch: the person enters an agentic state, stops feeling like the author of their acts, and executes orders almost automatically. Obedience is passive — a kind of moral sleepwalking. This is the classic textbook story, and it makes the participants sound like machines.
Engaged-followership reading
People obeyed because they identified with the cause — a scientific study they believed was worthwhile — and were persuaded, not commanded. Compliance rose when the experimenter invoked the value of the research and fell when he simply issued orders (Haslam & Reicher, 2017). Obedience is active: troubled followers doing what they think is right for a leader they trust.
The ethics firestorm
Milgram's paradigm didn't just teach us about obedience; it helped teach the field how to treat human participants. The objections were serious and, by today's standards, decisive. Participants were deceived about the study's purpose, the reality of the shocks, and the identity of the "learner." There was no meaningful informed consent, since no one could consent to a procedure they weren't told about. Participants showed real psychological distress — sweating, trembling, nervous laughter, and in some cases seizures of anxiety. Their right to withdraw was not merely unstated but actively denied: the whole design hinged on an experimenter insisting "you have no other choice, you must go on." And some participants left with inflicted insight — unwelcome knowledge about their own capacity for harm that they had never agreed to acquire.
Diana Baumrind's 1964 critique in American Psychologist landed hardest: she argued the study lacked informed consent, that the emotional harm was real, and that Milgram reported that harm in a detached tone that clashed with the vividness of his own descriptions of participants' distress. Milgram defended himself with follow-up data — reporting that 84% of participants were glad to have taken part and that a psychiatrist found no lasting trauma — and with the argument that the results justified the discomfort. But Perry's archival work later showed that Milgram's debriefing was itself misrepresented in print: many participants were not properly dehoaxed and did not learn until much later, if ever, that the shocks were fake (Perry, 2013). Modern IRB requirements — genuine consent, minimized harm, a real right to withdraw, thorough debriefing — are in no small part a response to studies like this one.
Bodhi says
Don't let the ethics debate and the reappraisal collapse into "so it was all bunk." That's the lazy conclusion. The right posture is two-handed: the study was ethically indefensible by modern standards and methodologically messier than its legend and the core effect is real and replicable. Holding all three at once is what critical thinking about classic research actually looks like. Anyone who tells you Milgram was either flawless or fraudulent is selling you a cartoon.
The companion cautionary tale: Zimbardo's prison study
Milgram usually travels with a partner — Philip Zimbardo's 1971 Stanford Prison "Experiment," in which students randomly assigned to be guards or prisoners in a mock basement jail seemed to descend, within days, into cruelty and breakdown. For decades it was taught as the ultimate demonstration that toxic roles corrupt ordinary people. It's worth reading with a critical eye, because it has not survived scrutiny nearly as well as Milgram's paradigm. Historians and journalists working from Zimbardo's own archived tapes found serious problems: it was a demonstration, not a controlled experiment, with no comparison condition; guards were not left to "become" cruel but were coached toward harshness in an orientation briefing; at least one prisoner's famous "breakdown" was later described by the participant as acting; and the whole setup was saturated with demand characteristics — everyone knew what story the researchers wanted. Thibault Le Texier's 2019 analysis of the archives makes the case in detail (Le Texier, 2019). The contrast with Milgram is instructive: one paradigm bent under re-examination but held; the other largely came apart. We track both verdicts on the Science Corrects Itself page.
The one thing to carry out of this topic
Milgram's real lesson isn't "people are sheep." It's that ordinary situational forces — a legitimate authority, tiny increments, diffused responsibility, no time to think — can pull decent people into indecent acts, and the same situational levers can just as easily set them free. The obedience rate wasn't a fixed fact about human nature; it was a dial that Milgram spun, and one dissenting peer could turn it down to almost nothing. Carry that dual insight — the danger and the exit — along with the mature habit of holding a classic study's real contribution, its ethical failures, and its methodological limits all in view at once. That's the whole topic.
References
Baumrind, D. (1964). Some thoughts on ethics of research: After reading Milgram's "Behavioral study of obedience." American Psychologist, 19(6), 421–423. https://doi.org/10.1037/h0040128
Burger, J. M. (2009). Replicating Milgram: Would people still obey today? American Psychologist, 64(1), 1–11. https://doi.org/10.1037/a0010932
Haslam, S. A., & Reicher, S. D. (2017). 50 years of "obedience to authority": From blind conformity to engaged followership. Annual Review of Law and Social Science, 13, 59–78. https://doi.org/10.1146/annurev-lawsocsci-110316-113710
Le Texier, T. (2019). Debunking the Stanford Prison Experiment. American Psychologist, 74(7), 823–839. https://doi.org/10.1037/amp0000401
Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4), 371–378. https://doi.org/10.1037/h0040525
Milgram, S. (1974). Obedience to authority: An experimental view. Harper & Row.
Perry, G. (2013). Behind the shock machine: The untold story of the notorious Milgram psychology experiments. The New Press.