Dissecting Eyewitness Memory and Why Confidence Can Be Misleading

October 12, 2021 By Jane Kelly, jak4g@virginia.edu Jane Kelly, jak4g@virginia.edu

A new Showtime documentary series on the groundbreaking case of a father sentenced to life in prison based on the repressed memories of his daughter is bringing the issue of eyewitness memory and its reliability back into national conversation.

In 1990, a jury handed down the life sentence to George Franklin. This was after his daughter, Eileen, said memories of the grisly murder of her then-8-year-old friend, Susan Nason, flashed in her mind while gazing into her young daughter’s eyes.

The dark tale is being played out again on the cable giant’s new series, “Buried.”

Chad Dodson is a University of Virginia professor of psychology who leads a lab that focuses on memory. While his work is not connected to the Franklin case, his research on eyewitness memory and its fallibility has bearing on today’s justice system.

Most recently, he and his team have been looking at eyewitness identification, especially when it is made with a high degree of confidence.

UVA Today reached out to Dodson to learn more about this research.

Q. Can you describe your research interests?

A. I’ve been at UVA for 20 years. I’ve always been interested in when people falsely remember past events and when people are highly confident about what they’re remembering, and yet they’re completely wrong.

Up until about five years ago, we looked at this in a variety of different ways and we looked at younger versus older adults. Then, about five years ago, we became interested in eyewitness identification. And we’re using that as a tool for examining when people make these high-confidence memory errors.

Related Story

For five years, Chad Dodson and his team have been investigating eyewitness memory and how overconfidence affects its veracity. (Photo by Dan Addison, University Communications)

Q. What is your definition of eyewitness identification?

A. By eyewitness identification, I’m really only talking about when eyewitnesses are shown a lineup of people, and the lineup may or may not contain the culprit. And I’m interested in when people are highly confident and yet wrong. So that relates to the more general issue of ‘When is confidence reliable as opposed to an unreliable predictor of accuracy?’

We’re interested in variables that seem to systematically predict that confidence-accuracy relationship – or in other words, variables that systematically predict when eyewitnesses may be making a high-confidence misidentification.

Q. How are you testing to determine if someone is making a high-confidence misidentification?

A. We’re not examining real eyewitnesses. We are not examining real crimes. Instead, we bring people into the lab. For a lot of our studies during COVID, we were doing this online.

In both cases, we show people a video of a mock crime, usually a robbery. And then after a delay of a varying amount of time, we then show people a lineup of faces and that lineup will either contain the person that they saw in the video or it will not. And we’ll ask people to either respond ‘Not present,’ [meaning] that nobody in the lineup fits their memory of who they saw from the video, or if somebody in the lineup does fit their memory, then we’ll ask them to choose that person and to give a confidence rating about the likely accuracy of their choice.

Q. How does the passage of time between seeing the mock crime and deciding about guilt affect the determinations of the study subjects?

A. We’ve tested thousands of people. There are a couple of variables that we look at.

One variable is how quickly or slowly it takes people to make an identification. So, if you choose somebody from the lineup, do you do that relatively quickly or does it take you a long time? And one variable that we’re finding that is a pretty reliable predictor of accuracy is how long it takes you [to respond].

We find that on average, those people who respond relatively quickly tend to be more accurate than those people who take a longer amount of time. And in fact, what I think is interesting is that the combination of decision time and confidence together, both of those variables are a very reliable predictor of accuracy. So, those people who are highly confident and they choose that person within a relatively short amount of time, their identification, on average, will tend to be highly, highly accurate.

Q. Have you determined precisely what amount of time it takes to make an accurate identification?

A. The $64,000 question is ‘How fast is fast?’ And this is something that is really hard to pin down. That’s why I’m talking about faster versus slower. But faster would be, you know, under five seconds. Slower would be maybe 30 seconds or longer.

Q. You’ve determined that the person who takes longer to answer is usually wrong. What do you infer from that and how can this information be practically used?

A. I’m interested in both the practical implications for law enforcement and ways of preventing false convictions and misidentifications, and then theoretically, from the memory point of view, we’re interested in what this tells us about the decision processes that people use.

So [our findings] suggests that those people who are taking a longer amount of time, they are perhaps using a looser criteria for evaluating their memory. They need perhaps less vivid information to say that they’re highly confident about their response, in which they may be wrong, whereas perhaps those people who are taking a shorter amount of time, those responses on average will tend to be more accurate. Perhaps they’re more accurate because those folks are using a stricter criterion for the kind of memorial information they need to identify somebody.

This is an ongoing line of research about what distinguishes slower identifications from faster identifications.

Q. How does facial recognition ability played into this research?

A. One variable that we’re really excited about – and this is sort of a kind of a new line of research – is face recognition ability. I think something that I hadn’t really appreciated until a couple of years ago is that there is a tremendous amount of variability, just from person to person, in how good of a face recognizer you are. Some folks are great; other folks are not so great.

After we give people this eyewitness paradigm that that I’ve just been describing, then afterward we’ve been giving people a standardized face recognition test that seems to be a pretty reliable measure of people’s face recognition ability. And we’re finding that your face recognition ability is a decent predictor, not surprisingly, of your accuracy. But it’s a decent predictor of whether you’re likely to make a high-confidence misidentification. So, folks who are just below-average face recognizers tend to be much more likely to make high-confidence misidentifications than folks who are above average.

Q. Subjects of your research are either given a delay of a few minutes or a day before being asked to make an identification. What percentage of people make misidentifications?

A. In conditions where there’s a one-day delay, on average, people will overall be wrong more often than if there’s just a short five-minute delay. On average, in the five-minute delay, maybe people are correct … 60% of the time; and then after a one-day delay, on average, people will be correct … roughly 30% of the time.

Then, within that average, there’ll be a lot of variability, depending upon whether you’re a strong face recognizer or weak face recognizer. And also depending upon how quickly or slowly you made the identification or whether you made the identification with high confidence or low confidence. So, that average point hides a lot of interesting information.

Q. How can this information be used in the justice system?

A. Practically, we know that confidence is the most persuasive factor at influencing juror decision-making. … Eyewitness confidence is the most persuasive factor. So, jurors who hear an eyewitness say that she or he is highly confident that that was the person, that was the culprit, that is very, very influential.

So, one way of reducing false convictions is educating law enforcement and jurors about when high-confidence identifications may not actually be accurate and may be inaccurate.

Media Contact

Jane Kelly

University News Associate Office of University Communications