Milgram's Obedience Experiments

In brief

In 1961, months after Adolf Eichmann went on trial in Jerusalem, the Yale psychologist Stanley Milgram (1933 to 1984) recruited ordinary residents of New Haven for what he advertised as a study of memory and learning. It was not. Each volunteer was told to deliver increasingly severe electric shocks to another person every time that person answered a question wrong. The shocks were fake and the "learner" an actor, but the volunteer did not know that. Before running the study, Milgram asked psychiatrists and colleagues to predict the results; they expected almost no one to reach the maximum. In the first published condition, 26 of 40 participants, 65 percent, did. The experiment became one of the most famous, most replicated, and most contested findings in psychology.

The full treatment

The question it set out to answer

Milgram had worked as a research assistant to Solomon Asch (1907 to 1996), whose experiments showed people would conform their perceptual judgments to a group's obviously wrong answer. Milgram wanted to push further: not whether people would echo a false opinion, but whether they would actively harm another person on command. The Eichmann trial gave the question urgency. Hannah Arendt, covering the trial for The New Yorker and later in Eichmann in Jerusalem (1963), described the defendant not as a monster but as a colorless bureaucrat who insisted he was only doing his job, coining the phrase "the banality of evil." Milgram opened his own book with this exact puzzle: could ordinary Americans be made to act the same way, simply because someone in authority told them to?

How the experiment worked

Volunteers answered a newspaper ad for a study on learning, paid a small fee, about $4.50, before the session started and told the money was theirs regardless of what happened, to remove money as a reason to comply. A rigged drawing always assigned the real volunteer the role of "teacher" and a confederate the role of "learner," strapped into a chair in an adjacent room with an electrode on his arm. The teacher sat before a generator with thirty switches, labeled in fifteen-volt steps from 15 to 450 volts, running from "Slight Shock" through "Danger, Severe Shock" to a final, unlabeled "XXX." For every wrong answer the teacher moved one switch higher. The learner's recorded protests escalated with the voltage, banging on the wall around 300 volts, then falling silent. If the teacher hesitated, an experimenter issued one of four fixed prods in sequence, from "Please continue" to, finally, "You have no other choice, you must go on." The session ended only when the teacher refused outright or reached 450 volts three times running.

What the results showed

Participants were not calm about it. Milgram described volunteers sweating, stuttering, groaning, and digging their nails into their palms, some breaking into fits of nervous laughter, while most kept pressing the switches. In the baseline voice-feedback condition, 65 percent went all the way to 450 volts. Milgram's own explanation was the "agentic state": under a legitimate-seeming authority, a person shifts from seeing himself as an autonomous moral agent, responsible for his own acts, to seeing himself as an instrument carrying out someone else's will, responsible only for performing the role well. In that state, the moral weight of the act attaches to the person giving orders, not the hand on the switch.

The variations that mapped the limits

Milgram did not stop at one condition. Across more than a dozen variations, obedience moved with the social architecture of the situation, not with any fixed trait of the teacher. Putting the learner visibly in the same room dropped obedience to roughly 40 percent; forcing the teacher to press the learner's hand onto a shock plate dropped it to roughly 30 percent. Moving the experimenter out of the room, giving orders by phone, collapsed obedience to about 20 percent. When two confederate co-teachers refused and walked out, obedience fell to about 10 percent. But when the participant's only job was reading the word list while a confederate threw the actual switch, obedience rose to roughly 92 percent. A plain Bridgeport office instead of Yale's halls lowered obedience somewhat but did not erase it. Proximity to the victim, distance from the authority, and visible peer defiance moved behavior more than personality ever could.

Related distinctions

Obedience, in Milgram's sense, is compliance with a direct command from a perceived authority. Keep it distinct from Asch's conformity, the pull to match a group's judgment with no command involved, and from Philip Zimbardo's Stanford Prison Experiment (1971), which studied how an assigned role, guard or prisoner, reshapes behavior with no one giving orders at all. Keep it distinct too from the normative question treated in the entry on authority: Milgram asked whether people obey, not whether they should.

Lineage

The experiments descend from Solomon Asch's conformity paradigm and answer directly to Hannah Arendt's account of Eichmann, turning a claim about bureaucratic evil into a testable design. They fed forward into Zimbardo's Stanford Prison Experiment and into the broader situationist current in social psychology that also produced Lee Ross's work on the fundamental attribution error, all pressing one claim: circumstance shapes conduct more than character does.

The strongest case for it

The core behavioral finding has proven durable. Thomas Blass's meta-analysis of replications conducted between 1961 and 1985, across several countries, found obedience rates that varied by setting but showed no systematic decline over time, undercutting the idea that Milgram merely caught a uniquely deferential postwar generation. Jerry Burger's 2009 partial replication, stopped at an ethical threshold of 150 volts, found continuation rates statistically indistinguishable from Milgram's own, forty-five years earlier. A virtual-reality version by Mel Slater and colleagues in 2006, using a computer-generated learner participants knew was not real, still produced measurable stress. The finding also does real work: it gives empirical teeth to Arendt's claim that catastrophic harm needs no monster, only an ordinary person inside a structure that reroutes responsibility upward, and it launched an entire research tradition on why good people do harmful things.

The strongest case against it

The objections arrived quickly and have never fully receded. Diana Baumrind's 1964 critique, published within a year of Milgram's first paper, argued that the study inflicted real, foreseeable psychological harm through deception no debriefing could fully repair, and that no participant could be said to have given informed consent to an experience designed to make them believe they might be injuring or killing someone. That critique, more than any other single paper, helped push American psychology toward the ethical review boards and consent requirements that now govern human research.

Gina Perry's 2013 archival investigation, working from Milgram's own recordings and unpublished files at Yale, found the tidy published account considerably messier. Experimenters often departed from the fixed four-prod script, improvising extra pressure or reassurance. Debriefing was inconsistent, and some participants left believing, longer than Milgram's reports suggested, that they had badly hurt someone. Perry also found many voiced doubt, during or after the session, about whether the shocks were even real, which complicates any claim they knowingly overrode their conscience.

The deepest challenge questions what the obedience actually was. S. Alexander Haslam and Stephen Reicher, reanalyzing the wording of the four prods, found participants would generally continue after the first three, which frame continuing as necessary for science, but would almost always refuse after the fourth, a bare command with no justification attached. Their reading: participants were not blindly submitting to authority but persuaded, with reasons, that their actions served a valuable scientific goal, "engaged followership" rather than mindless obedience. If so, the parallel to bureaucratic atrocity weakens, since the historian Christopher Browning, in Ordinary Men (1992), found German police reservists ordered to shoot civilians could and did opt out, unpunished, far more often than Milgram's design would predict.

Where it stands now

The behavioral result, that a majority of ordinary volunteers will escalate harm under a credible authority's direction, has survived sixty years of replication and remains one of the most secure findings in social psychology. What has not survived intact is the simple story built on top of it. Milgram's agentic-state account has lost ground to identity-based readings in which the most obedient are those who came to see the authority's project as their own, and the ethical fallout of the design, more than the design itself, reshaped how human-subjects research is conducted everywhere. The experiments are taught less as proof that people are sheep and more as a case study in which levers, distance, legitimacy, peer dissent, move behavior toward or away from harm.

Test yourself

Picture yourself as the teacher, hearing the pounding on the wall and the silence after it, with a calm voice beside you saying the experiment requires you to continue. Guess, honestly, how far you think you would go. Then remember that the psychiatrists Milgram surveyed guessed wrong by an enormous margin, and ask what exactly you are relying on when you feel certain your own guess is not the same mistake.