Let’s Get Rid of Police Lineups

But what should we do instead?

Posted Sep 30, 2019

Getting rid of police lineups sounds like a crazy idea.  After all, there are plenty of crimes where police investigators rely heavily, and sometimes exclusively, on the testimony of eyewitnesses.  And we know that when an eyewitness picks the police suspect from a lineup that particular piece of evidence usually plays a critical role in securing a conviction because jurors are strongly swayed by positive identifications.

But, when witnesses are asked to view a lineup—either a live lineup or, more commonly, an array of photos—to see if they can identify the person whom they saw commit the crime, there exists an overwhelming body of scientific evidence demonstrating that they often make mistakes. And this pattern prevails even when best practice recommendations for lineup composition and conduct are observed.

Sometimes an innocent suspect is picked, with devastating consequences for the suspect (as highlighted by the US Innocence Project) and, of course, the very real possibility that the culprit will remain at large. Often, witnesses don’t pick anyone from the lineup, even though it contains the culprit. Or they might pick one of the known-to-be-innocent “fillers” (i.e., the people who are placed in the lineup with the suspect). This too may result in the culprit remaining undetected.

Some researchers have suggested that the problem can be largely overcome if the justice system just relied upon identifications that are made with extremely high confidence at the time the lineup is conducted—because highly confident identifications are likely to be accurate and low confidence identifications denote an unreliable memory.

As I pointed out in a previous commentary, however, some researchers have questioned whether we should rely on the “very high confidence = very high accuracy” conclusion when evaluating the likely guilt of a suspect in an individual case.  Moreover, as we have argued in a paper that has just been published in the journal American Psychologist, very high confidence identifications—that is, identifications made with 90-100% confidence—are not particularly prevalent.

So what makes lineups (either live or photo arrays) so problematic? The lineup requires a witness to make a categorical—or all or none—decision: the witness either picks someone or they do not make a choice (e.g., they say ‘not there’ or ‘don’t know’).  Unfortunately, this categorical decision is shaped not only by the quality of the witness’s memory but also by characteristics of the individual, their reflections on how the crime and identification test characteristics might have affected their memory, and social influences that shape how much “memorial” evidence they require to make a positive identification decision. Consequently, for a number of years, I and my colleagues have argued that the categorical lineup decision is an insensitive index of the match between the lineup stimuli and the witness’s memory.

We have shown in a number of studies that a much more sensitive index of that match and, in turn, the likelihood that the police suspect is the culprit, can be obtained by (a) asking witnesses to indicate their confidence (on a 0-100% scale) that each of a number of possible candidates, one of whom is the suspect, is the culprit, and then (b) using the resulting pattern of confidence judgments to predict the likelihood that the suspect is indeed the culprit.

Using this approach, our recent American Psychologist paper provides compelling demonstrations of the likelihood the culprit is in the array under a variety of conditions—evidence that is simply not obtainable using the traditional identification testing approach. For example, when witnesses gave the maximum confidence rating to the suspect, rather than to one of the lineup fillers, the probability that the suspect was guilty was very high when that maximum value was high. And, although probability of guilt fell as that maximum value fell, probative information about suspect guilt was still available even when the maximum value was quite low. The latter finding is really important because, in a traditional lineup situation, a witness who judged the suspect as the one most likely to be the culprit, but was not particularly confident in that judgment, would be most unlikely to choose from the lineup. In other words, information of probative value would be lost from a substantial proportion of witnesses.

Notably, the findings just described also applied to a sample of children as well as to adults.  This is a very important finding because it has been well documented that lineup identifications from children are of questionable accuracy. When the culprit is in the lineup they perform at similar levels to adults; but, unfortunately, children have strong tendency to choose when presented with a lineup which means that the likelihood of error is high if they are presented with a culprit-absent lineup.

If, however, one of the fillers in the array, rather than the suspect, received the maximum confidence rating, it indicated the suspect was more likely innocent than guilty. In other words, if the witness thought someone other than the suspect was the best match to their memory, it was a clear pointer to the suspect’s likely innocence. 

Another striking and perhaps counterintuitive pattern was observed. Sometimes witnesses could not split two members of the array: that is, although the suspect was given the maximum confidence rating, one of the fillers also received the same rating. When using a traditional lineup it seems unlikely that the witness would make a positive identification under such circumstances  Yet, it turned out that, when using our new approach, probability of suspect guilt was high if the suspect received a high confidence rating, even if that maximum rating was shared with another array member. Moreover, probative information was still obtained from witnesses even when the shared maximum values were not extremely high.

The overall picture to emerge from the quite intricate response patterns described in the American Psychologist paper is that witnesses obviously can access a rich array of information in memory that points to the likelihood of a suspect’s guilt or innocence. Yet, much of this information simply does emerge when using the relatively crude—and prone to error—police lineup that has been around for a very long time, but simply does not align with contemporary understanding of how recognition memory works.

What then of the future? From a research perspective, we need continued work to refine our understanding of the potential contributions of the approach we have outlined. From the perspective of the justice system, the promise of our “confidence” procedure affords an opportunity to do away with the highly problematic police lineup and the associated, but flawed, connotations of definitive suspect guilt or innocence. In its place, the approach we have outlined provides a form of identification evidence that parallels other forms of scientific evidence: namely, a probabilistic indication regarding the likelihood that the police suspect is the culprit. Such evidence may be hard for the police and the courts to get used to, but it reflects the realities of what eyewitnesses can tell us.


Brewer, N., Weber, N., & Guerin, N. (in press, 2019 online first).  Police line-ups of the future?  American Psychologist.  DOI: 10.1037/amp0000465