Is eyewitness identification more reliable than we think?
By Kirsten Weir
February 2016, Vol 47, No. 2
Print version: page 40
Weir, K. (2016, February 1). Mistaken identity. Monitor on Psychology, 47(2). https://www.apa.org/monitor/2016/02/mistaken-identity
One night in 1984, a stranger broke into Jennifer Thompson-Cannino's apartment and raped her. After the assault, Thompson-Cannino, then a 22-year-old college student, helped police sketch artists create a composite picture of her attacker. Later, in a photo lineup, she identified Ronald Cotton — a 22-year-old man who looked strikingly like her sketch and had previous run-ins with the law. Then, she picked Cotton from a live lineup. Cotton was convicted of rape and sentenced to life in prison.
A decade later, DNA testing revealed that Cotton was not a match to semen samples from Thompson-Cannino's assailant. But the samples did match the DNA of another convict, Bobby Poole — who, it turned out, had told a fellow inmate he had committed the crime.
In the years since Cotton was exonerated, he and Thompson-Cannino co-authored the 2010 bestselling book "Picking Cotton" about their experience, and have campaigned together to reform eyewitness identification procedures.
Their story has become another classic example of the fragile nature of eyewitness testimony. Beginning in the 1990s, forensic DNA testing has revealed hundreds of cases of wrongful convictions. In fact, eyewitness misidentification has played a role in more than 70 percent of wrongfully convicted individuals, according to the Innocence Project, an organization that works to exonerate wrongfully convicted people.
Following that flurry of DNA-fueled exonerations, law enforcement agencies started paying closer attention to the science of memory and identification, says Gary Wells, PhD, a psychologist at Iowa State University who has studied eyewitness identification since the 1970s. Police departments across the country began making changes to lineup procedures, such as presenting possible suspects one at a time rather than all at once. In 2011, the New Jersey Supreme Court issued a landmark decision requiring judges to instruct jurors on the limits of eyewitness identification. Other states are discussing similar instructions, Wells says.
"I honestly believe there's no area of experimental psychology that has had a bigger effect on the legal system," Wells says.
Now, however, some psychologists say some of those changes may have been premature. "What has been described as a success story of psychological research is not looking like it's quite so simple anymore," says Steven E. Clark, PhD, a psychologist at the University of California, Riverside. "Eyewitness identification may be more reliable than we think."
Much of the recent debate involves lineups — or, more commonly, photo arrays. In a traditional lineup, a witness views six to nine potential suspects (or their photographs) simultaneously. But in the 1980s, Wells started testing a new method: sequential lineups, in which possible suspects are viewed one at a time rather than all at once. He found that sequential lineups resulted in a significant decrease in the false identification rate, with only a small drop in the correct identification rate.
In the years since, numerous studies have supported that finding. A 2011 meta-analysis of 72 studies by Wells and colleagues found that eyewitnesses made fewer mistaken identifications when suspects were presented in sequential, rather than simultaneous, lineups (Psychology, Public Policy, and Law, 2011). In light of such evidence, the Innocence Project endorsed the method, and many police departments changed their lineup procedures.
By some estimates, around a third of law enforcement agencies in the United States now use the sequential format, says John Wixted, PhD, a psychologist at the University of California, San Diego. But, he says, that switch might have been a mistake.
Wixted is one of several scientists, along with Clark and Scott Gronlund, PhD, a psychologist at the University of Oklahoma, who have championed a statistical method called receiver operating characteristic (ROC) analysis, a method widely used in other fields to measure the accuracy of diagnostic systems.
Using that analysis, sequential lineups don't appear to be beneficial — and might lead to slightly more misidentifications than simultaneous lineups, Gronlund and Wixted have reported (Current Directions in Psychological Science, 2014). The problem, they say, is that previous analytical methods confounded accuracy with a witness's willingness to choose a suspect. In other words, sequential lineups seem to make people less likely to make a choice at all. But when they do pick a suspect, they might be at greater risk of making the wrong choice. "It turns out sequential lineups are inferior," says Wixted.
Unlike previous analytical methods, ROC analysis takes into account how confident a witness is at the time he or she picks a suspect from a lineup. And confidence, it turns out, is worth paying attention to.
For many years, researchers didn't think an eyewitness's confidence revealed much about his or her accuracy in identifying a suspect, says Wixted. A confident eyewitness could be just as likely to get the ID right — or wrong — as a less confident witness. But in the last two decades, numerous analyses have converged on the fact that eyewitness confidence is actually a strong indicator of accuracy.
"On issues of both confidence and lineup format, the field for years was making a mistake in how they analyzed the data, and in both cases arrived at the wrong conclusion," Wixted says. "And in both cases the incorrect conclusion was transmitted worldwide, including through police departments all across the nation — and that's where we stand today."
Wells remains skeptical of the ROC method. "It works in certain contexts, but not in a lineup," he says. However, having advocated for sequential lineups in years past, he now says he suspects that in the end, the two lineup methods won't differ much in their rates of false identifications. Of all the variables that influence the accuracy of eyewitness identifications, he says, the sequential-versus-simultaneous factor is one of the least important (Law and Human Behavior, 2015).
Much more critical, Wells says, is the choice of "fillers," the nonsuspect individuals in a lineup. Fillers should generally fit the witness's physical account of the suspect. "If you use poor fillers, an innocent suspect who does fit the description is at high risk of being identified," he says.
Wells says one of the riskiest situations for mistaken identifications occurs when the real suspect isn't present. "People have incredible difficulty recognizing the absence of the perpetrator," he says. "If he's there, they can do a fair job of picking him. But if he's not there, they tend to pick somebody else."
Yet law enforcement officers can include anyone in a lineup. To minimize that risk, Wells recommends creating administrative rules that require some reasonable suspicion before presenting someone's face to the witness. "Any time you put an innocent person in a lineup, there is inherent risk that they might be identified. But it turns out there are no rules for detectives to go by."
When Thompson-Cannino identified Cotton in court, she was certain she was pointing to the man who raped her. And yet she was wrong.
As the Innocence Project statistics make clear, many eyewitnesses are mistaken, even confident ones. So how is it that confidence can predict an accurate identification?
The problem, experts say, is in considering what witnesses say on the stand. "We shouldn't even ask witnesses to make an ID in the courtroom because that's really kind of worthless," Gronlund says. "What did they say two months ago when they first saw the guy?"
When Thompson-Cannino was first shown photos of possible suspects, she spent several minutes deliberating between two candidates. When she finally chose Cotton, she said, "I think that's him." But over time, her certainty grew.
Many people think of memory as a kind of mental YouTube; when you want to recall an event, just find the right link and hit play. In reality, memories aren't merely played back but are freshly reassembled every time they're recalled. That makes them extremely susceptible to modification.
Imagine a witness who hesitates when identifying a suspect from a lineup. The police officer smiles and says "good job," and the witness's confidence grows. Later, the witness sees the suspect's photograph in the newspaper. A lawyer asks, "Do you remember if the suspect had pierced ears?"
What emerges is an inaccurate memory that is rich with details that feel very real, says Elizabeth Loftus, PhD, a psychologist at the University of California, Irvine, well-known for her work on human memory. "The original memory is fading, you're reconstructing the event with new information available to you, or new sources of suggestion. You imagine something, and it feels like a memory."
For that reason, the only eyewitness identification that really matters is the first one, Wixted says. "You don't get to go back six months later and examine a crime scene again because it's been contaminated by then. Memory works the same way," he says. "You get one memory test, and one test only."
Wells agrees it's crucial for law enforcement officers to pay attention to initial confidence. In a recent meta-analysis of studies from around the globe, he and his colleagues found that confirming feedback (such as "Good, you picked the suspect") strongly inflated an eyewitness's confidence in his or her judgments (Psychology, Public Policy, and Law, 2014).
"You have to treat eyewitness evidence as a form of trace evidence," Wells says, not unlike the fibers and hairs and gunpowder residue that investigators collect from a crime scene. "It's easily contaminated, and it can fade with time."
In 2014, a committee of the National Academy of Sciences issued a report recommending best practices for eyewitness identification procedures. The committee's report underscored the importance of cluing into a witness's initial confidence.
Among the committee's findings was a recommendation that law enforcement officers videotape the initial witness-identification process and document the witness's confidence judgment at the time he or she first identifies a suspect.
The committee also recommended that lineups and photo arrays be double-blind, so that any officer administering the test doesn't know who the true suspect is. Such a move would go a long way toward preventing officers from influencing eyewitness identifications — inadvertently or otherwise — and inflating witness confidence, says Loftus.
One area the National Academy committee did not take sides on was the debate over sequential versus simultaneous lineups. The report's authors concluded that previous analytical methods conflated accuracy with willingness to make an identification. However, despite concluding that ROC analysis has advantages over previous methods, the authors also noted some concerns that ROC could be influenced by certain measurement errors and biases.
Ultimately, the committee called for using a broader array of statistical tools to tackle the question, and recommended that law enforcement agencies stick with their existing protocols until the science is clear.
Wixted agrees with the academy's recommendation for more research. "I think it's essential that the larger community of basic experimental psychologists turns its attention to these matters," he says.
Although there are lingering disagreements among eyewitness identification researchers, those in the field generally agree with the National Academy recommendations, and they welcome the focus that eyewitness research is being given.
"Despite the fact that there are a few issues and controversies, we still have a whole lot of things we agree on," Loftus says. "In recent times we've seen some great successes, in terms of the legal system waking up and paying attention to the body of scientific work. As scientists continue to refine their methods, I hope these points of controversy won't take away from the rest of the story."
In Clark's view, debates in the field are a good thing. "A science is really healthy when we take stock of what we don't know, and ask new questions or ask old questions in different ways," he says. "The science is going to be much better as a result."
Still, Clark cautions that it may not be possible to significantly reduce the number of mistaken IDs without letting more criminals go free. "One of the claims that the eyewitness field has made for years is that we can implement new procedures that will substantially reduce the false identification rate with little or no loss of correct identifications," he says. "If you look at the literature, this is just not true."
To be sure, there's a widely accepted view in the law that it's better to let the guilty go free than to punish an innocent person. But to make the most informed decisions about eyewitness procedures, Clark says, law enforcement officers and policymakers need to be aware of the trade-offs.
Wells, however, is hopeful that science will continue to chip away at the problem, finding new and better methods to reduce mistaken IDs without reducing accurate ones. "The legal system wouldn't be able to function without eyewitness identification," he says. "We have to find ways to make it better."
