The science of why eyewitness testimony is often wrong
The advent of DNA testing has made it uncomfortably clear that our criminal justice system often gets things wrong. Things go wrong for a variety of reasons, but many of them touch on science, or rather the lack of a scientific foundation for a number of forensic techniques. But in 70 percent of the cases where DNA has overturned a conviction, it also contradicted the testimony of one or more eyewitnesses to the events at issue.
According to a new perspective published in PNAS, that shouldn’t surprise us. The paper’s author, Salk neuroscientist Thomas Albright, argues that we’ve learned a lot about how humans perceive the world, process information, and hold on to memories. And a lot of it indicates that we shouldn’t value eyewitness testimony as much as we do. Still, Albright offers some suggestions about how we can tailor the investigative process to compensate a bit for human limitations.
Persistence of memory
Albright has some history in this area, as he co-chaired a study group at the National Academies of Science on the topic. His new perspective is largely a summary of the report that resulted from the group, and it’s an important reminder that we have sound, evidence-based recommendations for improving the criminal justice system. Failure to implement them several years after the report is problematic.
According to the perspective, things go wrong for eyewitnesses right from the start. While human vision is good, there are plenty of conditions—low lighting, distance, and sudden actions—that make it difficult to accurately perceive what’s going on. And we don’t always focus on the things that would make us a good eyewitness; if someone’s waving a gun around, we tend to look at the gun, not their face. And even if they’re not waving a gun, we may believe that they were if the visual information we recall is ambiguous, but we know there was a robbery happening.
Faced with partial information, research indicates that our brain’s response isn’t to commit what information we do have to memory. Instead, our brains attempt to create a coherent picture that makes sense. This often involves filling in details using past experience as a guide. The resulting memory may be satisfyingly complete, but it can come at the cost of incorrect information.
Memories are also remarkably malleable. Rather than remaining tucked away unchanged in our hippocampus, lots of research shows that the mere act of recalling something can allow its memory to be updated (or, in some cases, lost). And there’s a good chance that witnesses will be asked to recall events many times before they’re finally called upon to testify in court. Often, one of those times includes viewing a lineup, either in person or via photos. This can leave a witness open to suggestions made by investigators who already have a suspect in mind—suggestions that could then be used to influence the memory.
By the time someone gets to testifying at trial, their confidence in their recall of the relevant events isn’t actually based on whether they can accurately remember them. Instead, it has mostly shifted to a confidence about whether they can piece together a coherent picture of what happened, which isn’t the same thing. (Albright quotes psychologist Dan Kahneman as saying, “Declarations of high confidence mainly tell you that an individual has constructed a coherent story in his mind, not necessarily that the story is true.”) Yet for a jury, the confidence of the witness tends to be a critical part of evaluating a witness’ credibility.
That isn’t to say that no eyewitness’ memory is reliable or our judicial system’s reliance on their testimony is completely misguided. Instead, Albright echoes the National Academies report in arguing that the judicial system needs to do more to take the limits of memory into account. And police departments can implement procedures meant to limit their role in influencing a witness’ recall.
Albright cautions that the research in some areas is still in flux. For example, an early study praised the use of sequential lineups, in which possible perpetrators are shown one by one, because it cut down on false identification. But follow-up work showed that the approach cut down on all identifications, false or accurate. Now, the pendulum has swung back toward recommending simultaneous viewing.
That said, some of the changes that the report recommends focus on the key point in the process: the initial identification in a police lineup. It calls for making the procedure “blind,” meaning that the police running it are not involved in the case and have no idea about who the potential suspect might be. The procedure should be videotaped, and the witness’ confidence in the identification should be ascertained at the time. (Albright cites a case where a person who was uncertain at a lineup later went on to express complete certainty in their mistaken identification in court.)
The courts can also play a role through the instructions they give to both witnesses and juries. Witnesses can be instructed in ways that help provide a consistent and conservative response and include a sense of uncertainty. Juries can be informed of some of the limits of eyewitness testimony.
Albright is somewhat optimistic about the potential for change, citing two recent state court decisions that improved the scientific foundations for allowing eyewitness testimony. But he notes that for federal courts, the Supreme Court has set a standard for eyewitness testimony back in 1977, and it was based on legal precedent rather than science. There are no obvious means of updating that standard.
But ultimately, the most significant recommendations in the report have to be implemented at the level of thousands of individual police departments across the country. While courts can help drive that process, the change will only be possible if these departments value accuracy over a possibly inflated chance of getting any sort of identification.
PNAS, 2017. DOI: 10.1073/pnas.1706891114 (About DOIs).
The science of why eyewitness testimony is often wrong