Obedience: How Ordinary People Follow Orders to the Edge

Here is the first thing the famous number leaves out: it is the result of a single condition. Milgram did not run one experiment; he ran at least nineteen, and by later tallies as many as twenty-one to twenty-four variations, and obedience swung wildly depending on the setup. When the learner sat in the same room, full obedience fell to around 40 percent; when the teacher had to physically force the learner's hand onto a shock plate, to about 30 percent; when the experimenter gave orders by telephone instead of in person, to about 21 percent; and when two other 'teachers' (also actors) refused and walked out first, obedience collapsed to around 10 percent. Move the whole study from Yale to a nondescript downtown office and it dropped too. A 2014 pooled analysis of twenty-one of Milgram's conditions, covering 740 participants, put the overall obedience rate at 43.6 percent, with condition-by-condition results ranging from roughly 10 percent to over 90 percent. The lesson is not that a fixed two-thirds of humanity will kill on command. It is that obedience is exquisitely sensitive to the situation, to how close the victim is, how legitimate and present the authority seems, and whether anyone else is seen to resist. The 65 percent is real, but it is the reading from one carefully arranged room, not a constant of human nature.

The famous number has a second complication, and it comes from Milgram's own filing cabinets. The writer Gina Perry, working from his archived tapes, notes, and correspondence for her book 'Behind the Shock Machine' (2013), reported that the reality was messier than the published account. On the recordings, many participants can be heard doubting that the shocks were real, and Perry argues that those who were told unambiguously that the shocks were dangerous tended to stop, while the standard, publicized procedure had the experimenter repeatedly reassure them that the shocks, though painful, were not dangerous. The experimenter, she found, frequently went off Milgram's own four-prod script, improvising extra pressure, and Milgram appears to have tuned the procedure across earlier pilot runs until he arrived at the high-obedience version that became famous. These are Perry's documented conclusions from the primary sources, not settled consensus, and they should be held as her reading. But they matter, because they mean the clean 65 percent was in part engineered. Separately, scholars still argue about why obedient participants obeyed at all. Milgram's own theory was an 'agentic state,' a switch by which a person stops feeling responsible and becomes a mere instrument of the authority. A powerful modern alternative, Alex Haslam and Stephen Reicher's 'engaged followership,' built from those same Yale archives, holds nearly the opposite: many participants were not switched-off automatons but active, willing collaborators who had come to believe they were advancing important science, and continued because they identified with the cause, not because they had stopped caring. And the core effect does seem durable: a careful 2009 partial replication by Jerry Burger, stopped for ethics at the 150-volt mark, found obedience there statistically indistinguishable from Milgram's, nearly fifty years on.

No account of obedience is complete without the study most often taught beside Milgram's, and the honest news about it is that it does not hold up. In 1971 Philip Zimbardo assigned 24 young men to be 'guards' or 'prisoners' in a mock prison in a Stanford basement; the guards turned cruel, the prisoners broke down, and the planned two-week study was stopped after six days. For decades it was presented as proof that ordinary people spontaneously become tyrants once handed unchecked power. That reading is now largely discredited. Thibault Le Texier's 2019 examination of Zimbardo's own archives showed the guards were not left to discover cruelty on their own: they were briefed beforehand on the specific psychological states the researchers wanted to produce and coached toward harshness, in effect handed a script. Zimbardo ran the study while also playing its 'prison superintendent,' a compromised dual role; there was no control group, a tiny sample, and no standard measures. Its single most iconic moment, the breakdown of 'Prisoner 8612,' Douglas Korpi, was later described by Korpi himself as a performance, staged so he could leave and get back to studying for exams. And a careful 2006 replication, the BBC Prison Study, failed to reproduce the descent into tyranny at all. The Stanford Prison Experiment remains culturally influential and is still widely taught, but it is not reliable evidence for the claim that made it famous. It belongs in this article as a cautionary tale, about experiments as much as about prisons.

This subject, more than most, tempts two opposite and equally wrong conclusions. The first is the dark one: that these studies prove virtually anyone will become a torturer, or a killer, the moment an authority commands it. The evidence does not support so sweeping a claim. Obedience ranged from roughly 10 percent to over 90 percent across Milgram's own conditions; even in the notorious baseline, 35 percent of people refused to go to the end; and the Stanford study, which is usually enlisted to make the point most vividly, cannot support it at all, because its cruelty was coached rather than spontaneous. What the work really shows is situational vulnerability under specific, identifiable conditions, not a fixed law that anyone will do anything if ordered. The opposite overclaim is the reassuring one, now fashionable in the wake of the critiques: that the studies are so tainted, tuned, staged, or seen through that they show nothing at all. That also goes too far. Asch's conformity effect has replicated across 133 studies in 17 countries; Burger's modern partial replication found obedience essentially unchanged after half a century; and even the sharpest critics of Milgram, Perry and Haslam and Reicher included, do not conclude that no obedience effect exists. Their point is narrower and more useful: the single clean number and the tidy story oversimplify a real but variable phenomenon, and 'engaged followership' may explain it better than Milgram's own theory did. The honest position keeps the finding and discards the myth. People are far more swayed by situation and authority than they believe, and that is neither a life sentence nor nothing at all.