Independent New Hampshire Journalism · Northfield, NH

The Prison Experiment Was Coached. The Truth About Power Is Worse.

Everyone quotes the Stanford study to explain why decent people in uniform do indecent things. The archives say the guards were told what to do. That is a management problem, not a mystery — and New Hampshire just took away one of its cheapest ways to manage it.

For fifty years the story went like this. Take ordinary college students. Flip a coin. Half become guards, half become prisoners. Within days the guards turn sadistic and the prisoners break down, and the whole thing has to be shut down early because the situation swallowed everyone whole. Good people, evil place. Nobody’s fault. It is the single most quoted piece of psychology in every argument about police, prisons, and what a badge does to a person.

It did not happen that way. In 2019 the journal American Psychologist published an archival investigation by the French researcher Thibault Le Texier, who had gone through sixteen boxes of Philip Zimbardo’s own papers at Stanford, plus tapes, transcripts, and interviews with fifteen participants. What he found was not a study that got out of hand. It was a study that was steered.

That correction matters far past the psychology department, and it cuts the opposite way from how most people assume. If cruelty in institutions arrives on its own, like weather, then nobody is accountable and there is nothing to do but wince. If cruelty is taught, permitted, and rewarded by the people running the place, then it has a paper trail. It has supervisors. It has records. And records can be demanded.

What the archives show

Start with the guards, who were supposed to be the whole point. Zimbardo wrote for decades that they received no training and invented their own methods. The tapes from the orientation session on August 14, 1971 say otherwise. Zimbardo told the assembled guards the psychological state he wanted them to produce, ticking through boredom and frustration and fear, then arbitrariness, then the loss of privacy and individuality. He finished the list with a summary the guards could not misread: “We have total power in the situation. They have none.”

The undergraduate serving as warden had already run a version of this in a Stanford dormitory three months earlier. Eleven of the seventeen prison rules were copied straight from that student project. So were the middle-of-the-night counts and the work details and the whole daily schedule. The warden briefed the guards on which humiliations had worked the first time, down to the trick of filling a blanket with burrs so prisoners would spend hours picking it clean. Writing afterward, he described his assignment plainly: he was given the job of “trying to elicit ‘tough-guard’ behavior.” When one guard went soft on the third day, the warden pulled him aside and told him every guard needed to be a tough guard.

The guards, meanwhile, were not told they were subjects. Zimbardo folded them into the research team, gave them report forms, and let them believe the experiment was about the prisoners. Several later said so in writing. The most abusive guard of all, the one the prisoners nicknamed John Wayne, told a debriefing that he had decided a realistic performance would help the study: “I was an actor, and I was hamming it up.”

The data were thin, too. Of roughly 150 hours, fewer than fifteen percent were recorded. Nothing at all was collected on the third day. Zimbardo’s own graduate assistants warned him in internal reports that the video sample was skewed toward the dramatic and the unusual, which is precisely the material that makes a good slideshow and a bad measurement. Le Texier’s verdict is blunt: “an incredibly flawed study that should have died an early death.”

None of this is a fringe position, and it is not new. The psychoanalyst Erich Fromm pointed out in 1973 that two-thirds of the guards never committed a sadistic act, which is a strange result for a study about how easily situations turn people into monsters. In 1975 researchers showed that college students given only a description of the setup could guess the expected outcome at rates above ninety percent. In 1979 an Australian team ran three different prison regimes and found that a participatory regime produced guard-prisoner relationships that were, by their account, entirely civil. In 2002 Alex Haslam and Stephen Reicher ran a prison study for the BBC without installing themselves as the boss, and the guards did not spontaneously become tyrants. Every time somebody removed the coaching, the tyranny went with it.

The obedience experiments have their own version of this problem. Gina Perry’s archival work on Stanley Milgram, published in 2013, reports that Milgram ran more than twenty procedural variations with wildly different results, that in some conditions a clear majority disobeyed, and that his experimenter went well past the four scripted prods, pressing one subject twenty-six times. The famous figure everyone repeats came from one condition with forty men.

Why a rigged study was so useful

Ask why a broken experiment survived five decades of criticism and the answer is uncomfortable. It survived because it was useful, and it was useful because it excused people.

The situationist reading of the prison study says the room did it. Not the sergeant who set the tone. Not the chief who looked away. Not the town that never asked for the file. Zimbardo took that reading to Congress in 1971, to textbooks for a generation, and eventually into a military courtroom as an expert witness for one of the Abu Ghraib guards. A theory that converts an accountable chain of command into an impersonal force of nature will always find customers, and the customers are rarely the people at the bottom.

There is a second reason it stuck, which is that it feels true. Most of us have watched a reasonable person get a little authority and turn into somebody else. That instinct is not wrong. It is just imprecise, and the actual research on what power does is more specific and more useful than the myth it replaced.

What power really does to a brain

The measurable effects of power are not about turning good people evil. They are about attention.

In 2006 Adam Galinsky and colleagues published a set of experiments in Psychological Science in which participants primed to feel powerful were about three times more likely to draw a capital E on their own forehead facing themselves, unreadable to anyone across the table. In the same paper, high-power participants were worse at identifying other people’s facial expressions. Power did not make them cruel. It made them stop checking.

In 2014 Jeremy Hogeveen, Michael Inzlicht, and Sukhvinder Obhi went further, publishing in the Journal of Experimental Psychology: General. They primed people into high-, neutral-, or low-power states, had them watch someone else perform a simple hand action, and used transcranial magnetic stimulation to measure motor resonance, the automatic echo in your own motor cortex when you watch another person move. High-power participants showed less of it. The mirroring dimmed.

Keep the size of that claim honest. These are laboratory primes with modest effects, and psychology as a field has spent the last decade learning painful lessons about overreading exactly this kind of result. The prison study is the cautionary tale about what happens when a vivid finding outruns its evidence, and it would be an embarrassment to make the same error twice in the same article. What these studies support is narrow and still worth knowing: holding power reliably degrades the ordinary, automatic habit of tracking what other people are experiencing. Not conscience. Attention.

That is a far more actionable diagnosis, because attention can be forced from the outside. A camera forces it. A required written explanation forces it. A five-day deadline forces it. A neighbor with a copy of the file forces it.

Hubris is a job condition, not a personality

In 2009 the neurologist Jonathan Davidson and David Owen, a physician who served as Britain’s foreign secretary, published a paper in the journal Brain proposing what they called hubris syndrome: a pattern of grandiosity and contempt for advice, drifting toward loss of contact with reality and identification of self with state, acquired through the exercise of power rather than carried into it. They set out fourteen defining features, required at least three for the pattern to apply, and argued it tends to fade once the power is gone.

Two conditions in their account do the heavy lifting. It takes time in power, and it takes minimal constraint. Owen and Davidson noted that cabinets are poor brakes because their members owe their jobs to the person they would have to stop. They also made an observation that should be read slowly by anyone who thinks character is the variable: the parliamentary revolts that ended two of the prime ministers they diagnosed would not have happened had those leaders left office after eight years.

Sit with that. The proposed cure is not a better person. It is a shorter clock.

Pharaohs, dons, and the discipline of being watched

Run the pattern back through history and the same variable keeps showing up, and it is never virtue.

Pharaohs who answered to no living institution carved their names over their predecessors’ monuments and built colossi of themselves. Kings who claimed to embody the state governed like men who could not be contradicted, because they could not be. The behavior is remarkably consistent across three thousand years and every kind of throne, which is a strong hint that we are looking at the effect of a structure rather than a run of bad luck in the gene pool.

Organized crime makes the point with unusual clarity, because it is a pure test. Carlo Gambino ran the most powerful crime family in New York for roughly two decades. He was quiet, avoided publicity, discouraged the drug trade on the theory that long sentences make informants, and died of heart disease in his own bed on Long Island in 1976. His last stretch in prison had been in the 1930s. John Gotti took over the same organization in 1986, wore the suits, worked the cameras, and enjoyed the nickname. He was convicted in 1992 on thirteen counts, sentenced to life without parole, and died in federal custody in 2002. Both men were, by any moral measure, criminals. The difference in outcome tracked the difference in exposure. The one who behaved as though he were being watched lasted. The one who behaved as though he could not be touched did not.

Now apply that to a police department, which is the version that touches a person in Laconia or Gilford or Northfield on an ordinary Tuesday. An officer with a body camera that is on, a supervisor who reads the report, a prosecutor who must disclose credibility problems, and a resident who can request the file is operating in Gambino’s world. An officer whose footage is never released, whose discipline sits in a file the public cannot see, and whose department can run out the clock on a records request is operating in Gotti’s, minus the indictment.

Why we don’t elect the humble ones

Here is where this stops being about other people’s institutions and starts being about the choices we personally make in a voting booth.

The best evidence we have on this is a 2015 meta-analysis in Personnel Psychology by Emily Grijalva and colleagues, which pooled decades of studies on narcissism and leadership. The finding is clean and damning. Narcissism has a positive relationship with leadership emergence and no relationship at all with leadership effectiveness. Narcissists rate their own performance highly. The people who work with them do not. Effectiveness turns out to follow a curve with a moderate peak, not a straight line upward.

Read that again as a description of an election. The trait that predicts who gets chosen has no bearing on who does the job well. We are not screening badly. We are screening for the wrong thing entirely, and we do it because the selection happens in a few seconds of impression while the performance happens over years we mostly do not watch.

The money finishes the job. OpenSecrets reported that congressional races in 2024 cost about $9.5 billion including outside spending, and that even one of the safest House seats in the country still required its incumbent to raise roughly $1.5 million. There are three constitutional qualifications for Congress. The fourth, unwritten one is the ability to ask thousands of strangers for money without flinching, year after year. That is a personality filter, and it does not filter for humility.

So the people we would most want in office are the least likely to run, the least likely to win if they do, and the least likely to be believed when they say they are unsure. The wise, empathetic, effective leaders do exist. They are usually running a school district or a fire department or a select board, in jobs small enough that humility is an asset and the scrutiny is close enough to be real.

Which points at the only lever ordinary people hold. We cannot install better characters. We can shorten the clock, keep the file open, and make sure the person in charge is being watched by someone who does not work for them.

New Hampshire just removed a brake

New Hampshire’s Right-to-Know Law is one of the better tools in the country for exactly this, and it belongs to everybody equally. RSA 91-A gives every person the same access. There is no press tier and no media exception. If anything the statute cuts the other direction: the fee-waiver provision added in 2024 waives certain per-communication charges for indigent requesters and public-interest requests, and expressly carves the people who publish for a living out of that public-interest waiver. Whatever your politics, the law hands you the same key it hands a newsroom.

Enforcement is where the state has gone backward. In 2023 New Hampshire stood up a right-to-know ombudsman, a cheap and quick route for a resident to contest a denial without hiring a lawyer. Then the Legislature cut the office’s salary and spending budget to $30,000 and moved it under the newly created Office of State and Public Sector Labor Relations. Thomas Kehr, the first and only person to hold the job, resigned in July 2025. As the New Hampshire Bulletin reported on July 2 of this year, Gov. Kelly Ayotte has not nominated a replacement in the year since, and her office did not answer the Bulletin’s questions about why. The Secretary of State’s page for the office still carries the notice that no action can be taken on any filing until there is a new ombudsman.

Not everyone mourns it. William Chapman, a Concord attorney who has litigated open-records cases in this state for decades, told the Bulletin the ombudsman was “an unnecessary step” whose rulings were not binding anyway, and argued the better fix is to shorten agency response deadlines and force the courts to set hearings faster. He may well be right about the design. He is right about something more important: the pressure belongs at the front end.

Either way, the practical situation for a New Hampshire resident today is that the cheap door is bolted. What remains is RSA 91-A:7, which lets any aggrieved person petition the superior court for injunctive relief and, notably, appear with or without counsel. Under RSA 91-A:8 a body that knew or should have known it was violating the law can be made to pay attorney’s fees. That is a real remedy. It is also a courthouse, and most people with a denied records request are not going to a courthouse.

Police records sit right in the middle of this. It took a lawsuit brought by the New Hampshire Center for Public Interest Journalism, five newspapers, and the ACLU to establish that the state’s list of officers with credibility problems was not shielded as a personnel file. That list, once the Laurie List and now the Exculpatory Evidence Schedule, became a public record under RSA 105:13-d, enacted in September 2021, and the Department of Justice now publishes it. That is a genuine win. It also took a court fight, several years, and a coalition, which is roughly the opposite of a system that discloses on its own.

What works

Lord Acton’s line about power corrupting is quoted so often it has stopped meaning anything. The sentence directly before it, in his 1887 letter to Bishop Mandell Creighton, is the one worth keeping: “Historic responsibility has to make up for the want of legal responsibility.” When the law will not hold someone to account, the record does. That is the entire job description of a newsroom, and it is why a state that quietly defunds its records referee has done something more serious than trim a line item.

The prison study told a generation that the situation is stronger than the person, and that there is no point looking for someone to blame. The archives tell a better story. Somebody wrote the rules. Somebody handed out the schedule. Somebody told the quiet guard to be a tough guard, and somebody could have told him the opposite. Every one of those is a decision by a named human being, and named human beings can be asked to explain themselves in writing.

So stop waiting for humble leaders to rise. They mostly will not, and the research on how we choose explains why. Do the unglamorous thing instead. File the request. Note the date. Count the five business days. Demand the specific exemption when they withhold, because the statute requires them to name it. Go to the select board meeting where the body camera policy gets written, which is a room that is almost always half empty. And when your governor leaves the state’s records referee vacant for a year, treat it as the story it is.

Power does not need monsters. It needs an empty room and nobody taking notes.

— Dexter Dow, Granite State Report

Fact check

Editor’s note. Every factual claim above was checked against the sources named in the fact-check table before publication. The Le Texier paper, the RSA 91-A statute text, the Secretary of State’s ombudsman page and the New Hampshire Bulletin report were read in full. The Perry book, the OpenSecrets analysis and the Exculpatory Evidence Schedule are cited to their publishers and marked accordingly. Sources reporting different figures for the ombudsman office’s prior budget are noted in the table; only the figure all sources agree on appears in the article. No comment was sought from the Governor’s office for this piece, which is analysis rather than a news report on a specific pending action; the Bulletin’s reporting that the office did not respond to its questions is attributed to the Bulletin. Corrections: Granite State Report corrects verified errors promptly and appends a note identifying what changed and when.