False confessions: Not as rare as you might think

Today's article is one of my all-time favorites. It was originally written by Katherine Kiechel, an undergraduate at Williams College as part of her honors thesis, and could serve as a model for others in its simplicity and ingenuity (the report I'm discussing here was revised and coauthored by her professor, Saul Kassin: "The Social Psychology of False Confessions: Compliance, Internalization, and Confabulation," Psychological Science, 1996). Some empirical work has been done on false confessions, and at least one example of a genuine false confession has been reported: Paul Ingram, who confessed to the satanic slaughter of new-born babies, was later accused by a psychologist of a bogus crime. Though initially he denied it, eventually he "confessed," even fabricating new details about the supposed crime.

There are even manuals which advise police on the best way to extract confessions from suspects, which generally involve some level of deceit on the part of the interrogator. Though confessions extracted using threats or physical violence are typically disallowed by the courts, these milder techniques are generally accepted, and result in conviction and prison time for the accused.

However, Kassin and Kiechel were unaware of any experimental study of false confessions. Indeed, a study of false confessions for criminal behavior would probably be unethical, so they devised a task with significantly smaller stakes.

Participants—psychology majors at Williams College who participated for course credit—were told they were taking part in a study of reaction times. They were tested in groups of two, one of which was a confederate. The confederate read a list of letters to the naive participant, who then typed them into a computer as quickly as possible. Before the experiment began, participants were instructed not to hit the Alt key, as this would cause the computer to crash and data to be lost. Sure enough, about one minute into the experiment, the computer did "crash"—without the participant doing anything wrong—and the distraught experimenter rushed into the room and immediately accused them of hitting the Alt key.

Participants were then asked to sign a confession. If they refused, they were asked again. Sixty-nine percent eventually signed. Finally, when they left the experiment room, a second confederate was sitting in the waiting area, apparently ready to participate in the next session. The experimenter told this confederate she'd have to be rescheduled and left the room. To assess if they actually believed their confession, the second confederate asked the participant what happened. Twenty-eight percent of participants expressed belief in their guilt to this stranger, indicating that they had "pressed the wrong key" or something similar.

Finally, the experimenter brought the participant back to the original room and asked them to re-enact the mishap. This was done to see if people would confabulate, or invent, details supporting their confession. Nine percent of them did.

Kassin and Kiechel varied the conditions of the experiment in four ways. First, they made the reaction time task either slow and relaxed or fast and frenzied by using a metronome to time the letter-reading task (they had established earlier that a pace of 43 letters a minute was easy, but 67 was difficult). Second, they either had the confederate corroborate by falsely saying that she had seen the participant press the Alt key or truthfully say that she did not notice anything. This chart of results shows that these circumstances affect the results significantly:

The difference between having a corroborating witness or not was quite astounding. While 35 percent of those in the slow/no witness condition signed a confession, 100 percent of the fast/witness condition did. Even under the fast/no witness condition, only 12 percent internalized belief in their guilt by admitting it to the second confederate. Most shocking of all is the fact that in the fast/witness condition, 65 percent internalized their guilt, and 35 percent confabulated evidence. Since these are conditions that are quite similar—and possibly less extreme—to those advised in police handbooks for obtaining confessions, the rates of internalization and confabulation are strikingly high.

In a 2004 article in Psychological Science in the Public Interest, Saul Kassin and Gisli Gudjonsson report on follow-up research to the Kassin/Kiechel 1996 study, in which participants were told of punishments for confessing ranging from fines to 10 hours of free labor to repair the error. Despite these negative consequences, the researchers found similar effects. However, another follow-up found a much lower rate of confession when the key was the Esc key, a less plausible key to strike accidently.