Like three-quarters of the defendants who are cleared by DNA evidence, Cole was convicted based on sincere yet inaccurate eyewitness testimony -- a problem the New Jersey Supreme Court highlighted last week when it revised the state's rules for pretrial hearings and jury instructions based on three decades of research exposing the fallibility of human memory. That decision, together with a case the U.S. Supreme Court will hear this fall, reminds us that the most powerful testimony jurors hear may also be the weakest, subject to hidden influences that can send an innocent man to prison if they remain unexposed.
"Memory does not function like a videotape, accurately and thoroughly capturing and reproducing a person, scene or event," writes Geoffrey Gaulkin, the former judge who prepared a report on the science of eyewitness testimony for the New Jersey Supreme Court. "Memory is, rather, a constructive, dynamic and selective process" that includes three stages: acquisition, retention and retrieval. "At each of those stages," Gaulkin notes, "the information ultimately offered as 'memory' can be distorted, contaminated and even falsely imagined."
Tim Cole's wrongful conviction involved several of the memory-corrupting factors cited by Gaulkin. Cole's accuser, Michele Mallin (who later supported his posthumous exoneration), was held at knifepoint while she was raped, and the presence of a weapon can distract a victim's attention from facial features. High stress also impairs an eyewitness's ability to encode details.
Recommended
Police claimed that Cole resembled a composite picture they produced based on Mallin's description. "The broad consensus within the scientific community is that composites produce poor results," writes Gaulkin, "because people recognize others holistically, not feature-by-feature in the manner composites are constructed."
Mallin is white, whereas Cole was black, and cross-racial identifications are substantially less reliable than intra-racial identifications. Police presented Mallin with a photo lineup in which Cole's picture stood out because it was the only Polaroid and Cole was the only man looking directly at the camera. The detective who oversaw the lineup was conducting the investigation and may have unconsciously steered Mallin toward Cole, his only suspect.
Mallin was not told that her attacker might not be in the lineup or that she did not have to choose anyone if she was unsure. The photos were presented simultaneously instead of sequentially, encouraging "relative judgment" -- the tendency to pick the person who most resembles a perpetrator. Instead of recording the doubts Mallin had expressed, police made notes that suggested she had chosen Cole without hesitation.
The following day, the police did a live lineup, in which Mallin picked Cole again, reinforcing her initial misidentification. By the time of Cole's trial five months later, she confidently and convincingly identified him as her attacker, which was the only real evidence against him.
Gaulkin observes that courtroom identifications, for all their drama and emotional impact, merely recapitulate whatever errors led to an initial misidentification, hardened by feedback from police and prosecutors. He also notes that memory decays irreparably over time. According to a 2008 meta-analysis of 53 studies, memory quality falls by 50 percent after one month.
Because of these and other factors, eyewitness testimony is wrong about one-third of the time, judging from laboratory experiments, field studies and analyses of real cases. Almost all of the relevant evidence has emerged since the U.S. Supreme Court last addressed the due process implications of eyewitness testimony in 1977.
In November, the court will revisit the subject, considering whether eyewitness testimony can be suppressed when it is biased by factors other than "improper state action." However the justices rule, jurors in most cases still will have to decide not only whether eyewitnesses are honest but also whether they are telling the truth.
Join the conversation as a VIP Member