Split View: 밀그램 복종 실험 재검토 — 65퍼센트라는 숫자의 뒷면

The Milgram Obedience Experiment Reexamined — the Other Side of That 65 Percent

Introduction — Where Did the 65 Percent Come From

The summary always ends with this sentence. "65 percent of ordinary people delivered lethal electric shocks to a stranger."

The number itself is accurate. 26 out of 40, exactly 65 percent. The problem is that the fact that this number came out of a single condition is almost never mentioned. Stanley Milgram ran more than twenty variations between 1961 and 1962, and when the conditions changed, obedience dropped as low as 0 percent and climbed into the 90s. The point is not the number 65 but the range.

This installment of Psychology, Straight from the Papers is the companion piece to the Stanford prison experiment installment. The ending, though, is different. The prison experiment collapsed at the level of how the data were made, but the Milgram data have largely survived. What collapsed is the interpretation attached to them.

What the 1963 Paper Actually Did — 40 People, and 450 Volts

The 1963 Journal of Abnormal and Social Psychology paper by Milgram reports only the first condition.

• Participants were 40 men recruited through a New Haven local newspaper ad and a mailing. They ranged from 20 to 50 years old and came from a variety of occupations; none were Yale students. Payment was 4 dollars and 50 cents, handed over simply for showing up.
• The participant is always assigned the role of "teacher," and the learner role is played by a scripted actor. The draw was rigged.
• The shock generator had 30 switches running from 15 volts to 450 volts in 15-volt increments. The labels started at "slight shock," passed "danger: severe shock," and ended with two boxes marked simply "XXX."
• Whenever a participant tried to stop, the experimenter delivered four prods in a fixed order. "Please continue." Then "The experiment requires that you continue." Then "It is absolutely essential that you continue." Then "You have no other choice, you must go on."

The result is that famous 26. Nobody stopped before 300 volts, and 26 people went all the way to 450 volts. And the paper contains one more observation that popular summaries always leave out: the participants who obeyed were not calm about it. Milgram documented sweating, trembling, stuttering and fits of nervous laughter in detail. People did not obey placidly; they obeyed in the middle of intense conflict.

The Real Finding Is in the Variations — from 0 Percent to 92.5 Percent

Milgram went on to repeat the experiment while changing one condition at a time, and he pulled the whole set together in his 1974 book. The total number of participants runs well past 700. Here are the rates of reaching maximum voltage in the main conditions.

This table is the actual achievement of the Milgram research. Obedience was not a fixed human attribute but a response extremely sensitive to situational variables. It drops sharply as physical distance closes, falls to a fifth when the authority physically leaves the room, and nearly disappears when the person next to you refuses first. And it vanishes entirely the moment authority issues contradictory instructions.

Look especially at the last two rows side by side. When someone else pressed the switch, 92.5 percent stayed to the end. That means most cruelty happens not in the position of the person receiving orders but in the position where you never have to touch anything yourself, and this is a far more concrete and far more useful finding than "anyone can become a monster."

What the Yale Archive Revealed — Off-Script Pressure, Suspicious Participants

The Australian psychologist Gina Perry examined the original materials Milgram left behind — audio tapes, participant correspondence, experimenter notes — in person at the Yale library, and organized her findings in a 2012 book. Three things matter.

First, the experimenter did not follow the script. The paper states that the four prods were used in order and that the session was halted if the participant still refused after the fourth, but the tapes preserve scenes in which the experimenter presses the same person far more often than that, using words not in the script. This is a procedural deviation that may have raised the obedience rate above what it should have been.

Second, a substantial share of participants suspected the deception. This is not Perry alone making the claim; there are traces of it in the post-experiment surveys of Milgram himself. Even in the figures Milgram reported, only about half of participants said they "fully believed" that the learner was actually receiving painful shocks. If participants who did not believe it kept pressing the switch, then their behavior was measuring something other than obedience.

Third, what Perry emphasized was the problem of debriefing. A large share of participants did not hear the truth right after the session and went on for months or longer believing they had injured another person. This part leads into the topic of the next section.

An Alternative Explanation — Not Blind Obedience but Identification with a Cause

Here is where the reinterpretation by Alex Haslam and Stephen Reicher enters. Their starting point was a very concrete question: which of the four Milgram prods actually worked?

The fourth prod, "You have no other choice, you must go on," is the closest thing to a command. Yet the records show that this prod was the least effective of all. After those words, participants tended to stop instead. What worked well, by contrast, was the second prod, "The experiment requires that you continue." Not a command but a cause.

Haslam and Reicher gave this the name engaged followership. Participants did not do as they were told because they feared the authority; they continued because they felt they were collaborators on a worthwhile scientific project. This explanation fits the table above nicely. When two experimenters say different things, the cause is shaken and obedience goes to 0; when a peer refuses first, the legitimacy of the cause collapses and it drops to 10 percent. Run it in a commercial office and the cause of "Yale science" weakens, so it falls below half.

This reading has empirical support as well. Matthew Hollander and Jonathan Turowetz analyzed how participants in the original Milgram recordings explained their own behavior immediately after the session, and an overwhelming share of the explanations were not "I followed orders" but "I believed the learner was not really in danger" and "I was trying to help the research." The language of the participants themselves was not the language of obedience.

The 2009 Partial Replication — the Experiment That Stopped at 150 Volts

Once the ethics rules were in place, the original experiment could no longer be repeated as it was. Jerry Burger published a clever detour in American Psychologist in 2009.

• He seized on the fact that 150 volts is the branch point in the Milgram data. It is where the learner first protests and asks to be let out, and roughly 80 percent of the participants who pushed past it went all the way. So observing only up to 150 volts lets you estimate a good deal of the final outcome.
• 70 participants. Both men and women were included, a double screening filtered out applicants with psychological vulnerabilities, and participants were told at least three times that they could stop at any point. Debriefing was immediate.

The result: 70 percent tried to go past 150 volts. The comparable Milgram condition was 82.5 percent, and the difference between the two was not statistically significant. There was no difference between men and women either. Which means that 45 years had passed and society had changed a great deal, but behavior in this situation had not changed much.

This replication has clear limits too, of course. Because it stops at 150 volts, the genuinely contested 450-volt stretch is never observed, and the screened sample is more special than the original one. Still, at a minimum it is grounds for saying that the basic Milgram phenomenon is not merely a product of its era. A similar partial replication carried out by a Polish research team in 2017 produced a rate even higher than that.

The Legacy of Ethics Review, and What to Discard and Keep

The surest legacy of this experiment is not a result but a set of rules. In 1964, Diana Baumrind read the Milgram paper and published a head-on critique in American Psychologist. Her question was whether a researcher has the right to impose the cost of a participant learning that he is the kind of being who can harm another person. It would be an exaggeration to say that one Milgram experiment created institutional review boards, but it is clearly one of the cases most frequently cited while the norms for human-subjects research were being put in order through the 1960s and 1970s.

What to discard: the summary that says "65 percent of human beings will kill someone if they are just given the order." It is wrong in three places. 65 percent is the value from one of 24 conditions, close to half of the participants doubted whether the situation was real, and the actual behavior of the experimenter was more persistent than the script written in the paper. And the words the participants left behind were closer to cooperation than to obedience. Commentary that uses this experiment as grounds for explaining the cruelty of a particular group is equally unwarranted.

What to keep 1 — the variable is not the order but the cause and the distance. Look at the table again and you can see the intervention points. Close the distance to the victim and obedience halves; have one peer who refuses first and it nearly disappears. The single most powerful device for stopping a wrong instruction from being carried out inside an organization is the first dissenter, and there is 60-year-old experimental evidence for it.

What to keep 2 — the position where you never have to touch anything is the most dangerous. The 92.5 percent in the condition where someone else pressed the switch is the most uncomfortable number in this entire body of research. Automated systems, approval procedures split across many stages, and management structures that look only at metrics all manufacture that position structurally.

What to keep 3 — procedural transparency determines the lifespan of data. The reason the Milgram results held up longer than the Stanford prison experiment is not that the conclusion was more moderate but that the design changed conditions systematically and that he was in the habit of preserving raw materials. Even the criticism from Perry was possible only because Milgram did not throw away the tapes. Records that are left behind end up being used to refute the person who left them, and that is precisely the condition of good research.

A Guide to Reading the Originals

• Original paper: Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4), 371-378.
• Ethical critique: Baumrind, D. (1964). Some thoughts on ethics of research: After reading Milgram's Behavioral study of obedience. American Psychologist, 19(6), 421-423.
• All the conditions: Milgram, S. (1974). Obedience to authority: An experimental view. Harper and Row.
• Partial replication: Burger, J. M. (2009). Replicating Milgram: Would people still obey today? American Psychologist, 64(1), 1-11.
• Alternative reading: Haslam, S. A., Reicher, S. D., & Birney, M. E. (2014). Nothing by mere authority: Evidence that in an experimental analogue of the Milgram paradigm participants are motivated not by orders but by appeals to science. Journal of Social Issues, 70(3), 473-488.
• Recording analysis: Hollander, M. M., & Turowetz, J. (2017). Normalizing trust: Participants' immediately post-hoc explanations of behaviour in Milgram's obedience experiments. British Journal of Social Psychology, 56(4), 655-674.

Reading tip: the 1963 paper is short enough to read whole, but if you are pressed for time, start with the table in the results section (how many people stopped at each voltage band) and the paragraph describing the tension responses of the participants. Those two places are the real texture of this research. Then move to the summary table of obedience rates by condition in the 1974 book, and you will see at a glance why 65 percent cannot serve as a representative value. Finally, check the results figure in Haslam and colleagues 2014 to see which prod actually kept people going, and the heart of the interpretive debate falls into place.