Why innocent people admit to crimes

On 21 April 1989 a video camera starts running in New York. One after another, teenagers sit in front of it and describe what is supposed to have happened in Central Park two days earlier. They are fourteen, fifteen and sixteen years old. What the camera does not show: the long interrogations that led to these recordings were not recorded. The tape holds the result, not the route to it.

The statements contradicted one another about when, where and how the crime had taken place. None of the five described himself as a rapist; each said he had held someone down or watched while others acted. DNA recovered at the scene matched none of them. The police reported the finding as “inconclusive”. All of them retracted within weeks and said they had been intimidated, lied to and put under pressure.

They were convicted anyway, in 1990. They served between seven and thirteen years. Only in 2002 did the serial rapist Matias Reyes confess to the crime and state that he had acted alone; the DNA confirmed him. The verdicts were vacated, and in 2014 the City of New York settled with the five men for 41 million dollars.

Why would an innocent person confess? The question sounds reasonable and is dangerous for exactly that reason. It reconstructs a decision outside the situation in which it is made.

What the Reid technique demands

For decades one method has dominated in the United States, named after one of its authors: the nine-step Reid technique, set out in the manual by Fred E. Inbau, John E. Reid and colleagues. It is not built to stay open. It begins with an interview in which the investigator decides whether he believes the person in front of him is guilty. Only then does interrogation in the narrower sense begin — and it presupposes guilt.

Two families of tools carry the method. Maximisation stresses the strength of the evidence and makes every further denial look more expensive; it includes the evidence bluff, in which investigators claim that DNA, a video or a witness convicts the person when this is not true. Minimisation supplies a bearable story in return: it was an accident, anyone would have reacted that way, whoever talks now makes it easier on himself. The English PEACE model, by contrast, relies on open questions and gathering information rather than on a confession as the goal of the interview.

The difference is measurable. A meta-analysis using a network design evaluated 29 experiments with 81 effect sizes in which participants were questioned as suspects and the truth was known. Accusatorial interrogations produced considerably more false confessions than plain direct questioning (OR = 3.03; 95% CI 1.83–5.02) and more than the information-gathering approach (OR = 4.41; 95% CI 1.77–10.97). Conversely, the information-gathering approach yielded more true confessions than direct questioning (OR = 2.43; 95% CI 1.29–4.59). Almost all of the studies were run with students in the United States — the direction is better established than the magnitude.

What practice looks like has been reported by 631 investigators themselves. They put their own hit rate at detecting lies at 77% and the average length of an interrogation at 1.6 hours. By their own account they obtained self-incriminating statements from 68% of suspects — from innocent people in 4.78% of cases. The last figure in the survey is the striking one: 81% were in favour of recording interrogations.

Three routes to a false confession

Research distinguishes three patterns. Voluntary false confessions arise without immediate interrogation pressure, for instance out of a search for attention or a wish to protect someone. In the coerced-compliant confession the person knows they are innocent but gives way in order to end an acute strain or to obtain what appears to be a milder consequence. In the coerced-internalised confession they begin to doubt their own memory.

The third case sounds the least credible and has been reproduced best in the laboratory. Saul M. Kassin and Katherine L. Kiechel had participants work at a computer and warned them that one particular key must not be pressed. When the machine appeared to crash, they were accused of having done exactly that. The title of the paper names the three stages that followed: compliance, internalisation, confabulation. Some not only signed but afterwards remembered details of an act that had never taken place. Reconstructive memory fills gaps with what is plausible — and in the interview room someone else supplies the plausible.

To a guilty person an evidence bluff may seem predictable. An innocent person has to explain how impossible evidence can exist. Perhaps he blacked out; perhaps he remembers wrongly; perhaps the error can be cleared up later. And so trust in one’s own memory becomes the weak point.

What hours and sleep loss do

The most conspicuous quantity in proven cases is time. Steven A. Drizin and Richard A. Leo documented 125 false confessions; there the interrogations lasted an average of 16.3 hours. For comparison: in ordinary criminal proceedings the average length of questioning lies between 1.60 and 4.21 hours. Anyone who wants to explain a false confession therefore does not have to explain a stranger’s psyche, but a situation almost nobody has ever experienced.

How quickly exhaustion takes effect can be tested in isolation. In an adaptation of the computer experiment, participants spent a night either sleeping in the laboratory or awake throughout. In the morning all of them were asked to sign a statement that falsely attributed a rule violation to them. After a single request, the odds of signing were 4.5 times as high among the sleep-deprived as among the rested. One night is enough.

Added to this are characteristics that raise the risk: youth, intellectual or psychological impairment, isolation. In the Central Park case several of them came together — fourteen-year-olds, hours in custody, no lawyer.

How a confession rewrites the rest of the file

The decisive question is not only whether someone confessed, but what knowledge the confession contains. Does it hold accurate details of the crime that were not publicly known and not named beforehand by investigators? Or does the interrogation first introduce the information that later looks like the knowledge of a perpetrator?

This is precisely where the check regularly fails. Brandon L. Garrett examined 40 cases of demonstrated false confessions: in 38 of them the investigators claimed the suspects had named the decisive details on their own. In the Central Park case, officers took one of the teenagers to the scene before his recorded statement.

A confession also rarely stays on its own.

An analysis of the case files of DNA exonerations showed that in false confession cases several evidentiary errors came together more often than in cases of mistaken identification. In order of frequency, faulty forensic reports, eyewitness identifications, and snitches and informants accompanied the false confessions — and where several errors were present, the confession usually came first. It corrupts what follows it. That the DNA in the Central Park case incriminated none of the five and the finding was nevertheless passed on as “inconclusive” is not an exception but the pattern.

What remains

False confessions are not a marginal phenomenon: in around 30% of cases in which a conviction in the United States was later overturned, a false confession is among the contributing causes. What follows from this is unspectacular and has been the same recommendation for years. The entire interrogation must be recorded, from the first contact onward, not just the tidy closing statement. Only then can duration, breaks, claims about evidence and the origin of every detail be checked. For minors and particularly vulnerable people, legal and professional safeguards are central, and a confession without independently confirmed knowledge of the crime is not evidence but an assertion about an assertion.

How to recognise the problem in everyday life: by the question that gets asked. Anyone who wonders whether they themselves would ever confess to an act they did not commit is testing their convictions in daylight and after a night’s sleep. The question that counts runs differently. What would I believe after sixteen hours, if I were told there was evidence, and if this one sentence would end it all?