The Milgram Experiment Never Proved Blind Obedience

Here’s a fact I believed for most of my adult life: in 1961, a Yale psychologist proved that two out of three ordinary people will hurt a stranger simply because a man in a lab coat tells them to. It’s in the textbooks. It comes up in every conversation about how decent people end up doing terrible things.

Except that’s not what the experiment showed. I went down this rabbit hole expecting to confirm a scary fact, and came out the other side more unsettled — for a completely different reason.

This post is the written version of my video on the Milgram experiment — watch that one if you prefer crayon. If you’d rather read, stick around, because the paper trail here is genuinely strange.

A fake memory study at Yale

Stanley Milgram was a young psychologist who couldn’t stop thinking about the Eichmann trial — a Nazi bureaucrat who arranged mass deportations and then claimed in court that he was only following orders. Milgram wanted to test whether regular people could be pushed into harming someone by an authority figure. So he placed a newspaper ad in New Haven offering $4.50 for an hour of a “memory study.”

The setup was theatre. The volunteer played a “teacher” who read word pairs to a “learner” in another room, and punished wrong answers with electric shocks that climbed in 15-volt steps toward switches ominously marked XXX. The learner was an actor. The protests and cries coming through the wall were recordings. But the volunteer at the switchboard didn’t know any of that, and the strain was visible — sweating, nervous laughter, in a few cases something close to a breakdown. Milgram published the first results in 1963 as “Behavioral Study of Obedience”, and psychology has never really recovered.

The number everyone quotes

In the famous version, 26 out of 40 men went all the way to the final switch. That’s the 65%. Before the study, psychiatrists Milgram consulted had guessed maybe one person in a thousand would do it. You can see why the result exploded.

But here’s what most retellings skip: this wasn’t one experiment. Milgram ran more than twenty versions, quietly adjusting a single variable each time, and the number swung wildly. Bring the learner into the same room and full compliance fell to about 40%. Make the teacher physically hold the learner’s hand to a plate: around 30%. Have the experimenter deliver instructions by phone instead of standing there: roughly 20%. Let the volunteer watch two other people refuse first: about 10%. Even moving the whole thing from Yale to a shabby office building nudged the number down, which tells you the ivy on the walls was doing real work.

When researchers pooled all the conditions in the Meta-Milgram synthesis, the overall rate landed near 43% — and that’s an average of a dial that moved from 10 to 90 depending on the room. The 65% was the reading from one condition, at the top of the range. It became “the” number mostly because it was the scariest one.

Four sentences, and only one was an order

This is the part that rearranged my head. When a volunteer hesitated, the experimenter had exactly four scripted prompts, used in sequence: “Please continue.” Then, “The experiment requires that you continue.” Then, “It is absolutely essential that you continue.” And finally, “You have no other choice, you must go on.”

Notice that only the last one is an actual command. The first three are appeals — to the study, to the shared project, to being useful. And when researchers examined how each prompt performed, the pattern was backwards from everything the legend claims. Haslam, Reicher and Birney found the direct order was the one prompt that reliably failed. In the recordings, the moment people were told they had no choice, they demonstrated that they did — they quit. The bossiest line in the script was the least persuasive thing in the room.

Not obedience — recruitment

So if barked commands didn’t keep people at the switchboard, what did? The leading answer is what Reicher, Haslam and Smith call engaged followership: participants weren’t surrendering their conscience, they were siding with a cause. Science. Progress. Helping the nice institution finish its important work. The prompts that worked were the ones that made continuing feel like contribution, not submission.

There’s a darker wrinkle in the archives, too. One of Milgram’s assistants had privately rated how many participants truly believed the shocks were real, and that data sat unexamined for decades. When Perry and colleagues reanalysed it in 2020, the believers turned out to be the ones more likely to stop. The people who pushed to the end skewed toward those who suspected the whole thing was staged. Read that twice. The famous result partly measured doubt, not cruelty.

And before anyone relaxes: the pull itself is real. In 1972, Sheridan and King ran a version with a live puppy receiving genuine (small) shocks — no acting anywhere — and most participants still complied, in visible distress. Burger’s careful 2009 replication found around 70% continuing past the critical point, and a Polish replication published in 2017 reported nine in ten. There’s even brain evidence that acting on instructions dims your felt sense of authorship: Caspar and colleagues showed in 2016 that under orders, the brain processes your own actions more like things that merely happen.

The honest caveat

None of this comes from a tidy study, and I’d be cheating you to pretend otherwise. When the psychologist Gina Perry went through the original tapes for her book Behind the Shock Machine, she found the experimenter regularly drifting off script — improvising, pressuring, in one case pushing a single participant twenty-six times. Many volunteers went home without being told the shocks were fake, and some carried real guilt for months. Add the sample — 40 men per condition, one Connecticut town, the early sixties — and the ethics, which wouldn’t survive five minutes with a modern review board, and you get a result that’s fuzzy in both directions. Treat every Milgram number as a rough sketch, not a measurement.

What I’m taking from it

The lesson I grew up with was: beware the tyrant giving orders. The archive says something quieter and harder to guard against. Orders provoked rebellion. What kept good people pressing switches was a calm, reasonable voice assuring them the cause was worth it and their help mattered.

You’ll probably never face a shouted command to harm anyone. You will, at some point, hear a warm and sensible request to go along with something small that feels a bit wrong — for the team, for the project, for a good reason. That’s the moment this experiment was actually about.

If you like big questions taken apart honestly, one a week, subscribe on YouTube — it’s free, and it genuinely helps a small crayon channel keep going.