Stanford prison experiment or Stanford prison acting class?

4 minute read

I’ve previously written about the replicability crisis in science. I recently learnt about criticisms of the infamous Stanford Prison Experiment that I hadn’t seen when I was first reading about it. The criticisms invalidate the conclusions of the experiment for me, and I’ll highlight excerpts from this paper by Thibault Le Texier below:

Based on the first detailed published account of the experiment (Zimbardo, 1972a), Erich Fromm pointed out (1) the unethical nature of the harsh conditions imposed on the prisoners, (2) the fact that the personality pretests administered to the volunteers might not have detected a predisposition among some of the subjects for sadistic or masochistic behavior, and (3) the confusing situation for participants created by mixing realistic prison elements with unrealistic ones.

Not trying to downplay the importance of ethics, but that part is less interesting to me, as it doesn’t get to what I think is the core issue the experiment wanted to show, i.e. what does perceived power lead to in human behaviour? Selection bias and the lack of a fixed prison environment seem more interesting as a critique.

Banuazizi and Movahedi used a set of open-ended questions to determine the students’ thoughts as to what the experimenter’s hypothesis was and their expectations regarding the outcome of the experiment. 81% percent of the tested students accurately figured out the experimenter’s hypothesis (that guards would be aggressive and that prisoners would revolt or comply), and 90% predicted that the guards would be “oppressive, hostile, aggressive, humiliating” (p.158), thereby supporting the argument that demand characteristics were likely operating in the SPE and that the participants in the SPE would have probably guessed how Zimbardo and his co-experimenters wanted them to behave

This is more interesting! When you design an experiment you typically don’t want your subjects to know what you’re testing for, as they’ll likely game the system. E.g. if I tell you I’m testing whether you’re more or less likely to be rude after a perceived slight, you’ll conciously control or change your behaviour. Perhaps the students were just good at guessing? But we then see the below

The guards knew what results the experiment was supposed to produce. Zimbardo and his assistants announced the objectives of the experiment to the guards during their orientation day, Saturday, August 14, 1971. Zimbardo confided to the future guards that he had “a grant to study the conditions which lead to mob behavior, violence, loss of identity, feelings of anonymity. […]

This is convincing. If Zimbardo told the guards what he wanted to study and was looking for, no wonder results got so out of hand. And coupled with selection bias, perhaps that’s why the guards were so willing to act out their roles.

Far from reacting spontaneously to this pathogenic social environment, the guards were given clear instructions for how to create it

I’d say this is similar to the point above, that Zimbardo was essentially creating the scenario he wanted to observe. The paper raises two other reasons after this one that I’d say all fall under the ‘following Zimbardo’s instructions’ category, so I’ll skip that to go on to the next point

In order to get their full participation, Zimbardo intended to make the guards believe that they were his research assistants.

This also explains why the guards were so willing to go along. If the guards believed they were assistants that were helping conduct the experiment, no wonder they felt driven to get the results that Zimbardo was looking for. This leads to them playing their part like good actors:

During the debriefing that took place on the last day of the experiment, Eshleman confessed for example to Jaffe: “I thought it would be better for the study if I presented what I thought to be a realistic relationship between guard and prisoner. […] [T]hroughout the entire experiment I was an actor, and I was hamming it up. […] I was role playing.” (Eshleman, 1971a, p.2) Terry Barnett, the second most zealous guard, explained his involvement in the same terms.

It seems like the conclusion to draw from this is not that power corrupts, but instead that people will act in the way they feel incentivised to do so. As I’ve always written, incentives matter! In this case, the prison guards acting abusive to get the results they thought were needed, the prisoners staying and acting their part to get paid, and the researchers making the experiment seem more dramatic than it actually was.

Replicability has always been an issue in science, and I think we’ll see more such cases of ‘famous’ old studies that were just conducted badly, and whose results should be questioned. It’s a shame that there’s less focus on confirming old studies vs publishing your own new one, again due to incentive misalignment in the scientific field. As the author of the paper, Thibault, notes:

The SPE survived for almost 50 years because no researcher has been through its archives. This was, I must say, one of the most puzzling facts that I discovered during my investigation. The experiment had been criticized by major figures such as Fromm (1973) andFestinger (1980), and the accounts of the experiment have been far from disclosing all of the details of the study; yet no psychologist seems to have wanted to know if the archives what exactly did the archives contain.

and then speculates on a few reasons as to why:

Is it a lack of curiosity? Is it an excessive respect for the tenured professor of a prestigious university? Is it due to possible access restrictions imposed by Zimbardo? Is it because archival analyses are a time-consuming and work-intensive activity? Is it due to the belief that no archives had been kept? The answer remains unknown

I personally think it’s a combination of archival analyses being time-consuming ‘boring’ work and there being little incentive for researchers to do so.

If you like this, you might like my monthly newsletter