Heather D. Flowe, PhD
Eyewitness Identification from a Different Angle: Developing a 3D interactive lineup system
Updated: Jul 9, 2021
A single gunshot rings out as the double doors of your bank close behind you. There is a lone person running towards you. You get a closer look at him as he makes his getaway.
The tellers move swiftly in a choreographed fashion. Doors are locked, shades are drawn. The teller who was robbed is now writing, alone in an office. Police surround the bank within minutes.
Two weeks later, the police phone to ask you to see if you can identify the robber from a lineup:
Are you surprised? Your best view of the robber was of his profile, and yet, typical of lineups in many countries around the world, all the people are facing towards you.
On the one hand, people are said to possess extraordinary capabilities when it comes to remembering faces. There is even a special place in the brain (fusiform gyrus) to support these capabilities. Maybe you can extract a 3D representation of the faces and imagine how the faces look from all angles in your mind’s eye. In this way, perhaps you’ll be just as accurate as if you saw the lineup faces in profile.
On the other hand, much research suggests if you saw the faces in profile, you’d be better able to make an accurate identification. For instance, the encoding specificity principle holds that a high correspondence between the learning and testing environment in which memories are retrieved is a powerful determinant of remembering (Tulving & Thomson, 1973). Indeed, individual faces are better recognised if they are in the same angle as a face was studied. But the concept had never been tested in lineups, which contain the police suspect (who may be guilty or innocent) embedded among a number of other faces (called fillers). Testing the concept in a lineup matters because extensive validation is required by the legal system to ensure that any new lineup procedure meets evidentiary standards.
The more I thought about it, the more intrigued I became. And with that, the quest to develop and test a 3D interactive lineup system started.
An interactive system is born
I worked with programmers to develop the first interactive lineup system in 2015. Tests with the prototype system were indicating that a witness’ ability to discriminate guilty from innocent suspects could be significantly increased if witnesses were able to interact with the lineup faces.
Demo of the interactive lineup system
A system like this would be especially beneficial in the U.S. where most lineups are predominately static, frontal pose lineups, like the one the police showed you in the example above.
Before long I found myself at the headquarters of Promat to unveil the system. Lisa Smith was with me. She is a criminologist with expertise in developing new systems to be legally complaint. We had made the trip because Promat supplies UK police forces with lineup software.
We submitted our first funding proposal to conduct a controlled trial of the system with Promat as a partner. I also got to work developing the next generation of the software. The prototype version slightly distorted the images. This is no good because fidelity needs to be high in a court of law to establish identity. We also worked closely with the police to ensure that the system is both compatible with their requirements and cost effective.
Days before Christmas, while on holiday in Key West, I got the news that the proposal had been rejected. The two eyewitness identification reviewers rated the proposal as outstanding but the basic science researcher said, “Nothing new here, we already know this.” This was disappointing.
However, it forced me to re-think my approach. The work ultimately turned out better for it.
The most common factor associated with cases of mistaken identification is the witness being of a different race than the culprit. Faces are 1.4 times more likely to be misidentified if they are of a different race than the person making the identification. And other race identifications are ubiquitous in the legal system. Around 50% of IDs are cross-race identifications. This got me wondering, could an interactive lineup system improve accuracy in cross-race IDs?
Reducing Own-Race Bias
Own-race bias is the tendency to better remember faces that are of the same race as yours.
This is because people are more likely to undergo effortful elaboration and encode qualitatively diagnostic information for same- than other-race faces. Therefore, people are more likely to remember specific information that helps them better identify same-race faces compared to other-race faces.
The leading expert on own race bias is Chris Meissner. I reached out to him and thankfully he joined the team.
We thought that interactivity could perhaps boost performance because of the additional movement and viewpoint cues they provide. Chris and I received an award from the Laura and John Arnold Foundation to test this idea. Luckily, Melissa Colloff was finishing her PhD at the time and we were able to recruit her to the team. Melissa’s knowledge of modelling eyewitness memory performance is at the leading edge of the field.
So, what did we find? We tested more than 11,000 participants and found that interactive compared to static lineups significantly increased discrimination accuracy. You can read more about this work in our soon to be published in American Psychologist paper (the pre-print is here).
Why might interactivity be better?
We have conducted additional experiments to learn about the mechanisms behind better performance on interactive lineups. One idea is that interactivity boosts performance because witnesses can rotate the lineup faces and reinstate the pose in which they saw the culprit. We found evidence of this and submitted it for publication to Psych Science. We were stoked when the manuscript went under review. (Most submissions are desk rejected at this journal.)
Alas, I woke up one morning to find a rejection email and was more upset about it than when I was misdiagnosed with inoperable cancer some months before (I was/am totally fine). I would later laugh about this with Steve Lindsay, who was Editor in Chief of Psych Science at the time.
We carried on conducting experiments, and by this time, Melissa had become a Lecturer at Birmingham, and Travis Seale-Carlisle, a coding maven, was sent from above and joined the team as the postdoc. I’m pleased to say that the work on pose reinstatement has finally found a home and was accepted for publication in Nature’s Scientific Reports.
The experiments reported in the paper show that accuracy is higher when the viewing angle at learning and during the lineup match. The research also found that people spontaneously rotate the faces in the lineup to match encoding pose without any prompting from us. Thus, witness will actively seek out matching pose information during the lineup as an aide memoire, if they are given the opportunity.
We are continuing to investigate the effects of interactivity in a host of other studies. Melissa extended the procedure and tested children at the Think Tank Museum in Birmingham, in an award-winning exhibit that was designed and delivered by a team of UoB Psychology students and supported by the UoB Alumni Association. You can hear more about this research at SARMAC this month, where she will also deliver a keynote as the recipient of the J Don Read Early Career Prize. The paper is also in press at JEP: General (pre-print here).
My collaborator Harriet Smith spearheaded the first project examining the benefits of interactivity in forensic face matching, which is her area of expertise, both in mere mortals as well as in super recognizers. We recruited Josh Davis, who studies super recognizers, and David White, who is also an expert on forensic face matching and works with passport issuing teams in Australia, to the team. You can hear more about this work here and read about it in the British Journal of Psychology (preprint here).
We are also now examining the rotational behaviour of witnesses during interactive lineups. The system records how witnesses interact with the faces moment by moment. This information is being harnessed to profile accurate versus inaccurate identifications. Our students are also running studies that compare performance on interactive lineups to other types of lineup procedures used around the world, such as video lineups and simultaneous lineups.
Remembering back, the twists and turns along the way led to unexpected but exciting new pathways and collaborations. This has been an amazing project to work on, and I can’t wait to find out what’s around the corner.