A Prison That Was Not an Experiment

Two of the most cited studies in social psychology have turned out to rest on much less than everyone believed, and the reasons they collapsed are as instructive as anything they were supposed to show.

Start with Stanford, in 1971. Philip Zimbardo assigned student volunteers to be guards or prisoners in a mock prison in a basement, and the study was abandoned after six days because the guards had become abusive and the prisoners were breaking down. The conclusion, repeated in textbooks for fifty years, was that ordinary people placed in a situation with roles and power will enact those roles, and that the situation is sufficient.

Recordings released from Stanford's own archive show something different. The warden (a research assistant) instructed guards on how to behave, told them the prisoners needed to feel powerless, and pushed back on guards who were not being tough enough. One recording captures him coaching a reluctant guard directly.

A situation that includes explicit direction toward brutality does not demonstrate that situations produce brutality on their own.

The rest of the methodology does not hold up either. There was no control condition. The study ran six days. Outcome measures were observational and made by people who knew the hypothesis, including Zimbardo, who was simultaneously running the prison as superintendent. The volunteers had answered an advertisement specifying a study of prison life, and later work found that people who answer such an advertisement score higher on aggression and authoritarianism than those answering an otherwise identical advertisement without the word "prison". And one prisoner has said his famous breakdown was substantially performed because he wanted to leave and thought that was the way out.

The results were presented in a magazine before appearing in any journal, which is not incidental: the study's fame outran its scrutiny by decades.

The properly conducted version came in 2002, when Reicher and Haslam ran a prison study for the BBC with full ethical oversight, independent monitoring and no coaching.

The groups did not enact their roles. The guards, given no guidance about how to behave, failed to develop any shared sense of legitimate authority and were ineffective and divided. The prisoners developed a collective identity, organised, and eventually overturned the arrangement.

The conclusion is nearly the opposite of Stanford's, and more disturbing in its way. Tyranny is not what happens when you put people in uniforms. It requires people to identify with a group and to believe the group's cause justifies what they are doing, which is the same mechanism Haslam and Reicher found in the Milgram transcripts, and which cannot be prevented by avoiding uniforms.

Now the bystander effect, where the experiments are good and the founding story was not.

In 1964 Kitty Genovese was murdered in Queens. The New York Times reported that thirty-eight respectable, law-abiding citizens had watched from their windows and not one had called the police. The account horrified the country and launched a research programme.

Almost every element of it was wrong. Later investigation established that most neighbours heard rather than saw, that the attack happened in two separate episodes with most of the second out of sight, that several people did call the police, and that one woman went out and held her as she died. The reporter had taken a police official's offhand remark and built a story on it.

The research it launched, however, is sound. Darley and Latané staged emergencies and varied how many other people were present, and found intervention dropping reliably as group size rose. Two mechanisms: diffusion of responsibility (someone else will handle it, and if nobody does it is not particularly my fault) and pluralistic ignorance, in which each person looks at everyone else's calm to decide whether this is an emergency, and everyone is doing the same thing.

But the laboratory situations were deliberately ambiguous: smoke from a vent, a crash in another room. A 2019 study by Philpot and colleagues analysed CCTV footage of over two hundred real public conflicts in three countries, and found that someone intervened in around nine cases out of ten, and that more bystanders present made intervention more likely, not less.

So the effect is real and it is about ambiguity and unclear responsibility, not about crowds as such. In an unmistakable public assault, the crowd helps.

Which yields the one piece of advice from social psychology that fits in a sentence you can actually use. If you need help in a crowd, point at one person and tell them what to do. "You, in the blue coat, call an ambulance." Pointing assigns the responsibility that would otherwise diffuse; the specific instruction removes the ambiguity that makes people wait to see how others read it.

What survives from all of this is worth stating carefully, because the demolition of Stanford has been used to argue that situationism was nonsense.

The strong claim, that character is an illusion and behaviour is entirely contextual, has not held; personality measures predict behaviour aggregated across many situations reasonably well. What survives is the more modest point, which is still counterintuitive to nearly everybody: observers systematically underweight situational forces, and designing a situation is usually a more effective intervention than selecting for virtue.