A man in a grey lab coat says, "The experiment requires that you continue." The person at the switchboard has just heard a scream from the next room. He turns in his chair. He says he does not want to go on. The man in the coat repeats himself, flatly, as if reading. The person at the switchboard turns back and throws the next switch.
That is the scene most people carry from Stanley Milgram's obedience experiments, conducted at Yale between 1961 and 1962. In the baseline condition he reported in 1963, twenty-six of forty ordinary men from New Haven went all the way to the 450-volt switch, marked "XXX", on the instruction of an experimenter with no power over them. The learner, an actor, received no shocks. The participants did not know that.
The published result has been taught for sixty years as proof that ordinary people will do terrible things when told to. The archive of the experiments, held in Yale's Sterling Memorial Library, tells a more complicated story about how that result was produced.
Milgram kept almost everything: audio recordings of sessions, participant questionnaires, correspondence, notes on conditions he never published. The Australian psychologist Gina Perry spent years in those boxes and published her findings in 2012 as Behind the Shock Machine. Other researchers, including Stephen Gibson at York St John University and Nestar Russell, have worked through the same material.
Several things emerge.
The experimenter did not stick to the script. Milgram's published method describes four escalating "prods", ending with "You have no other choice, you must go on." The recordings show the experimenter, John Williams, improvising well beyond those four when participants resisted: arguing, reassuring, insisting, sometimes for minutes. The clean procedure of the paper was, in practice, an argument.
Debriefing was often delayed. Milgram wrote that participants were told the truth at the end of their session. The archive shows that many were not fully debriefed until much later, some receiving the explanation by post after the whole series was complete. In the meantime, a number of them believed they had hurt a stranger.
Many participants doubted the setup. Milgram collected data on whether people believed the shocks were real, and reported that most did. Perry and colleagues reanalysed an unpublished version of those data in 2020 and found that participants who said they fully believed the learner was being hurt were less likely to have obeyed to the end than those who expressed doubt. If that holds, some of the headline obedience came from people who suspected a trick.
There were many more conditions than the famous one. Milgram ran more than twenty variations. In several, obedience was low: when the experimenter gave orders by telephone, when two experimenters disagreed, when the participant had to press the learner's hand onto a shock plate. The 65 per cent figure is the peak of a range, not its average.
None of this was hidden at the time. Diana Baumrind published a critique in American Psychologist in 1964, the year after the first paper, objecting to the distress inflicted on participants and questioning whether laboratory obedience generalised to anything. Milgram replied that participants had reported being glad they took part. The archive shows some were, and some were not.
The behaviour itself has been reproduced. Jerry Burger at Santa Clara University ran a modified version in 2009, stopping the procedure at 150 volts, the point at which the learner first demands to be released. Roughly as many people continued past that point as in Milgram's baseline. The effect is not an artefact of 1961.
What has changed is the explanation. Milgram's own account was the "agentic state": people hand over responsibility to the authority and become instruments. Recent work has struggled to find that state.
The most telling evidence comes from the prods. Alex Haslam, Stephen Reicher and Megan Birney at the universities of Queensland and St Andrews showed in 2014 that the fourth prod, the only one that is a direct order, is the one people most often refuse. Burger, Girgis and Manning reported the same in 2011 from Burger's own replication: every participant who heard "You have no other choice" stopped. The prods that worked were the ones that appealed to the experiment's importance, not the ones that commanded.
Haslam and Reicher call this "engaged followership". Participants did not surrender their judgment. They were recruited to a cause, science at Yale, and kept going because they had come to believe in it. That reading fits the archive's picture of an experimenter arguing rather than ordering.
The strong version of the conclusion, that a lab-coated stranger can turn most people into torturers with a few flat sentences, does not survive the archive. The procedure was messier, the belief less complete, the obedience less uniform across conditions, and the mechanism something other than blind submission.
The weaker version survives well. Put an ordinary person in a situation with a legitimate-seeming authority, an escalating series of small steps, a worthy-sounding purpose and no obvious exit, and a large proportion will go much further than they would have predicted. That has been replicated, and it is enough.
It matters for the reader because the lab coat is rarely the point. What moved Milgram's participants was not the order but the story: this is important, this is science, you agreed to help. Anyone who has stayed in a bad situation because leaving would betray something they had signed up for knows that mechanism from the inside.