Confident eyewitnesses aren't necessarily more accurate, but a study of robbery cases finds that it depends on how much time has passed before you gauge their confidence.

Confident eyewitnesses aren't necessarily more accurate, but a study of robbery cases finds that it depends on how much time has passed before you gauge their confidence.


Eyewitness testimony may only be credible under these circumstances

In July 1984, a man broke into the apartment of Jennifer Thompson, a 22-year-old in North Carolina, and threatened her with a knife. She negotiated, convincing him to not kill her. Instead, he raped her and fled. Just hours later, a sketch artist worked with Thompson to create an image of the assailant's face. Then the police showed her a series of mug shots of similar-looking men.

Thompson picked out 22-year-old Ronald Cotton, whose photograph was on file because of a robbery committed in his youth. When word reached Cotton that the police were looking for him, he walked into a precinct voluntarily. He was eventually sentenced to life in prison based on Thompson's testimony. Eleven years later, after DNA sequencing technology caught up, samples taken from Thomson's body matched a different man who finally confessed. Cotton was set free.

When Thompson first identified Cotton by photo, she was not convinced of her choice. "I think this is the guy," she told the police after several minutes of hesitation. As time went on, she grew surer. By the time Thompson faced Cotton in court a year later, her doubts were gone. She confidently pointed to him as the man who raped her.

Because of examples like these, the U.S. justice system has been changing how eyewitnesses are used in criminal cases. Juries are told to discount the value of eyewitness testimony and ignore how confident the witnesses may be about whom they think they saw. Now, a new study of robbery investigations suggests that these changes may be doing more harm than good.

No one doubts that eyewitness evidence is risky. Hundreds of innocent people have been exonerated by DNA evidence in the United States since the 1990s, and eyewitness misidentification was the root cause in 70% of the cases. To get a handle on the problem, researchers have conducted experiments in which actors play out a crime, and subjects try to correctly identify the perpetrator from a lineup. The findings indicate that eyewitnesses can be easily fooled into confident false memories, and that showing them photographs sequentially reduces false identification. But many researchers have voiced concerns about whether these laboratory experiments accurately reflect real-world situations.

So a team led by John Wixted, a psychologist at the University of California (UC), San Diego, ran an experiment with the robbery division of the Houston Police Department in Texas. They focused on 348 robberies that occurred in 2013 and involved eyewitnesses and a single suspect. To make sure that the police could not unintentionally influence the process, officers who didn’t know who the real suspect was presented photos of six people to the eyewitnesses. When the eyewitnesses made a decision, they also noted how confident they felt about their choice on a three-point scale—high, medium, or low.

Each case also had a certain amount of corroborating evidence, ranging from weak—the suspect was known to be close to the scene of the crime—to strong—the suspect's shoes matched a footprint at the scene. If confidence really is a strong indicator of accuracy, the researchers reasoned, eyewitnesses who were more certain of their choice should be more likely to identify the suspects who also had abundant corroborating evidence, and eyewitnesses who double-guessed themselves should be less accurate.

And indeed, that's exactly how it worked out. Overall, the eyewitnesses fingered the suspect about one-third of the time, positively identified one of the five innocent people used to fill up the photo lineup one-third of the time, or decided that the perpetrator was not in the lineup at all one-third of the time. But factoring in the confidence of the eyewitness painted a different picture. Highly confident eyewitnesses fingered the suspect about 75% of the time, and falsely accused one of the five innocent people less than 20% of the time. And when the police had ample corroborating evidence against the suspects, the rate of positive identification by confident eyewitnesses shot up to 90%, the team reports online today in the Proceedings of the National Academy of Sciences.

The results don't mean that eyewitnesses aren't prone to confident false memories as in the Cotton case, Wixted says, but timing is key. "Our whole point is that you have to pay attention to confidence expressed on the initial identification only." Once an eyewitness gets to court to give testimony, false memories may have become crystallized.

In the majority of cases where DNA evidence has exonerated someone wrongly convicted, the initial eyewitness identifications were made with low confidence, not high confidence, Wixted says. If confidence scores were incorporated into the evidence, not only might fewer innocent people get accused but more guilty people might be convicted, he says.

The results of the study make sense, says Aaron Benjamin, a psychologist at the University of Illinois, Urbana-Champaign, because disregarding eyewitness confidence is "dramatically inconsistent with widely accepted views about memory." The justice system should reconsider how eyewitness testimony is used, he says, because "there's a lot at stake here." Courts may be "leaving more criminals on the streets and putting more innocent people behind bars than they should be."

But many researchers are calling for caution before changing the system. The study shows that there is real information to be gained from the initial confidence level, says Craig Stark, a psychologist at UC Irvine. But he worries that police, jurors, and even judges will misinterpret eyewitness confidence levels. In order for jurors to accurately weigh the value of testimony by its initial confidence, they will need to use sophisticated statistical reasoning, he says. "Can those subtleties be applied in a balanced way in people who are coming in with clear biases and preconceived notions?"