Reasons to Doubt the Reliability of Eyewitness Memory: Commentary on Wixted,Mickes,and Fisher (2018)

Wixted, Mickes, and Fisher (2018; this issue, p. 299) challenge the common trope that eyewitness memory is inherently unreliable (here, we use reliable and unreliable in the everyday sense, i.e., dependable and undependable). Recollections are indeed malleable, they acknowledge, and can be readily contaminated by improper interviewing techniques. But Wixted et al. assert that these facts should not lead us to characterize eyewitness reports as an unreliable form of forensic evidence. To support this assertion, they draw on mock-crime research and a small number of field studies, which indicate that high-confidence eyewitness reports are usually accurate, at least when memory is uncontaminated and suitable interviewing procedures are used. We agree with the thrust of Wixted et al.’s argument, and we welcome their invitation to confront the mass underselling of eyewitnesses’ potential reliability. Nevertheless, there is a comparable risk of overselling eyewitnesses’ reliability.

As we see it, Wixted et al.’s position is wholly compatible with the more common characterization of eyewitness memory as unreliable. The compatibility of these positions hinges on the base rate of “pristine conditions” (or at least near-pristine conditions wherein initial memory is uncompromised; Mickes, Clark, & Gronlund, 2017) in real-world eyewitness interviews. If near-pristine conditions are rare, then both the common trope and Wixted et al.’s characterization are plausibly correct.

Wixted et al.’s reasoning implies that near-pristine conditions or uncontaminated memories are normative, but we doubt this. To consider why, we must look to psychological science for answers to two questions. First, how often do eyewitness interviews deviate from best-practice and compromise the accuracy of witnesses’ reports? Second, how often have witnesses been exposed to preinterview influences that could corrupt reports obtained in best-practice interviews (see Poole & Lindsay, 1998)?

Answering the first question requires a definition of best practice. Wixted and Wells (2017) and many others (e.g., Levi & Lindsay, 2001; Wells et al., 1998) have sought to define the optimal conditions for eliciting identification decisions. Yet current evidence indicates that, despite promising shifts toward more widespread adherence, real-world investigators are not consistently following these best-practice guidelines. For instance, in a survey of more than 600 U.S. law enforcement agencies, Loftus and Greenspan (2017) found that some best-practice procedures are frequently met (e.g., instructions that minimize pressure on witnesses to make positive identifications), whereas others are met less than one third of the time (e.g., the use of double-blind lineups, explicitly probing witnesses’ confidence in their identifications).

Psychology is in a far weaker position to define the optimal conditions for eliciting eyewitness descriptions of people or events. To say this is not to ignore the substantial literature on best-practices in information-gathering interviews (e.g., Hope, Gabbert, & Fisher, 2011; Poole & Lamb, 1998). But we know far less about the relationship between accuracy and confidence under these different interviewing practices and less still about how well this relationship holds in real-world cases. Note also that the method used for calculating the confidence-accuracy relationship can affect the assessments made and the conclusions that are drawn. ¹ Unlike in the context of lineup identifications, we also have little evidence on how and when confidence should be measured when witnesses provide narrative descriptions. The practicality and effectiveness of assessing confidence on a detail-by-detail basis in real-world interviews is yet to be established.

Even if we could define and guarantee near-pristine interviewing conditions in the real world, intrainterview procedures are not the only possible sources of memory contamination. Wixted et al. note that “only the first memory test counts” (p. 327), but how often is the official, recorded report the first occasion on which the witness has talked about the event? We do not know the answer, but research shows that cowitnesses frequently discuss crime details before being interviewed (Eisen, Gabbert, Ying, & Williams, 2017; Skagerberg & Wright, 2008). Moreover, it seems likely that victims of sexual assault and domestic violence often talk to medical personnel, friends, or family before they talk to police (Starzynski, Ullman, Filipas, & Townsend, 2005). Such discussions can make it difficult to determine the source of crime-relevant details (e.g., Vredeveldt, Groen, Ampt, & Van Koppen, 2017) and can also create opportunities for memory contamination.

Lab studies show that people who receive misinformation are sometimes more confident in their erroneous memories than in their veridical memories (Assefi & Garry, 2003; Takarangi, Parker, & Garry, 2006), and they become increasingly confident in those erroneous memories as rehearsal makes them increasingly coherent and rich (Lindsay, Hagen, Read, Wade, & Garry, 2004). Moreover, individuals’ personal goals (Sharman & Calacouris, 2010), social goals (Hellmann, Echterhoff, Kopietz, Niemeier, & Memon, 2011), preferences (Gordon, Franklin, & Beck, 2005), expectations and stereotypes (Marsh, Cook, & Hicks, 2006), and semantic representations (Reyna, 1995)—to name just a few factors—can sometimes shape remembering even in the absence of misinformation. These influences typically occur without conscious awareness (Johnson, Hashtroudi, & Lindsay, 1993), and therefore justice professionals, jurors, and witnesses themselves can seldom know whether or to what extent such factors have contaminated a memory. In sum, even if eyewitness recognition and recall are reliable when uncontaminated, it is risky to depend on that knowledge without having reliable ways of detecting contamination.

When is “reliable” reliable enough? Wixted et al. claim that recall memory is highly reliable, citing evidence that when eyewitnesses report details with high confidence, they are highly likely to be correct. But is it appropriate to describe witnesses as “reliable” when around 10% of the information they report with “absolute certainty” in near-pristine conditions is incorrect (as in Odinot, Wolters, & van Koppen, 2009)? If witnesses in near-pristine conditions are prompted to answer questions only when they are sure, does it make sense to describe these witnesses as “reliable” when 9% of their answers are wrong (as in Evans & Fisher, 2011)? Wixted et al. are right to note that these witnesses were overwhelmingly accurate, and we agree that witnesses have been impugned by common claims that memory is unreliable. Moreover, many of the memory errors made by witnesses are likely to have little forensic relevance. But as Wixted and colleagues have acknowledged elsewhere, with error-rates of 10% or more, even high-confidence memory reports are “simply not a reliable enough indicator of truth to unilaterally adjudicate guilt or innocence” (Roediger, Wixted, & DeSoto, 2012, p. 113). Therefore to characterize eyewitness memory as inherently reliable could risk leaving judges, juries, and investigators with the unwarranted assumption that 100% confidence implies 100% accuracy. Indeed, one important difference between eyewitness memory and DNA is that lay decision makers are rarely left to assess the reliability of the latter for themselves (see Thompson & Newman, 2015).

It certainly seems timely to take stock of the robust and growing evidence on the relationship between accuracy and confidence and to avoid treating eyewitness memories as specious by default. But it would, in our view, be unwise to give the impression that eyewitness reports are inherently reliable.