The own-race bias in eyewitness identification

Catriona Havard, The Open University



The own race bias is the phenomenon whereby individuals are better at recognising and differentiating between faces that are the same race as they are, and much poorer with faces of another race.

The issue of the own-race bias has serious ramifications when it comes to eyewitness identification and when a witness is faced with the situation where they have to identify a culprit from a lineup who is of a different race. The innocent project organisation aims to reform the criminal justice system to prevent injustice has exonerated 315 people through the use of DNA evidence. 71% of cases involved eyewitness misidentification, the majority of which also involved a witness identifying a suspect from another race.

There are a number of theories as to why people are better at recognising a face from their own race. One theory, the contact hypothesis, argues that through the high level of contact that individuals have with own race faces, they become experts at recognising such faces (Brigham & Malpass, 1985; Slone, Brigham & Meissner, 2000). On the other hand, the comparatively lower amount of contact with other-race faces, leads them to be relatively inexpert at differentiating between other race faces (Hugenberg, Miller, & Claypool, 2007). According to the contact hypothesis, the more experience that one has with a different racial group the more accurate they should be at identifying members of that particular group (Brigham, et al., 2007).

In our latest research we wanted to see if the amount of contact children had with children of another race would influence how accurately they could identify a culprit of another race from a lineup (Havard, Memon & Humphries, 2017). In our study, we showed a group of Caucasian and a group of Asian children 2 mock crimes, one with a Caucasian thief and one with an Asian thief. After a delay of 1 or 2 days the children were shown 2 video lineups, one for each thief and asked if they could identify the culprits they had seen before.  Each child saw one line-up that contained one of the culprits that had been previously seen (culprit present) and one lineup that didn’t contain the culprit, but someone of a similar appearance (culprit absent). With culprit present line-ups, we were interested in whether the children could correctly identify a person from the line-up and if they were accurate with their own race. Whilst culprit absent line-ups,  were used to simulate the situation that the police have arrested the wrong person, and to investigate whether the children would still pick someone,  and make a false identification, even though the person they have seen previously is not there. We were also interested in whether children would make more false identifications for the culprit that was of another race. A measure of interracial contact was also taken, where children were asked about their contact and relationships with children of another race.

Our findings revealed an own race bias for the Caucasian children, this resulted in more correct identifications for the own race culprit from culprit present lineups, and more false identifications of the other race culprit for the target absent lineups. The Asian children from both age groups showed no own race bias and performed equally accurately for culprits of both races. The measures of interracial contact revealed that the majority of Caucasian children in our study had very little contact with Asian children, whereas the majority of Asian children had high levels of contact with Caucasian children. The more contact children had with children of a different race, the more likely they were to make a correct response when trying to identify someone of another race.

This article was originally posted on the OU Psychology blog at: To find out more about this research you can access the full article here or contact



How to improve identification evidence: Practitioner hits and academic false alarms.

Graham Pike and Virginia Harrison

The Open University

Previous articles written for this site have highlighted the dangers and limitations of eyewitness evidence (see Briggs and Westmarland, 2015), and described how such evidence can all too easily lead to a miscarriage of justice (see Kaye, Drake and Pike, 2014). For example, we noted the work of The Innocence Project, a largely US-based national litigation and public policy organisation, that has to date used DNA analysis to secure 330 exonerations of existing convictions. The average sentence served by the innocent people wrongfully convicted is 14 years, and 18 of the exonerations involved death sentences. Those stats might be frightening in themselves, but they are very likely the tip of a very large iceberg.

Embed from Getty Images
A great deal of academic research has been conducted on how and why eyewitness evidence can be so inaccurate, and academics and organisations like The Innocence Project have done their best to shout the results from the rooftops, yet very little change is evident in the gathering and use of such evidence within the criminal justice system. For example, other than a few localised pilot studies, recommendations coming from organisations such as The British Psychological Society and The American Psychological Association have been largely ignored.

In a recent piece of research conducted through the OU Policing Research Consortium and funded by the College of Policing, we explored why evidence from research concerned with witness identification is not being translated into practice. As a lot of research has already been conducted on the barriers to translation within organisations, including those of the criminal justice system, we instead focused on the research itself, looking at four specific areas:

  • The dissemination of research to criminal justice practitioners
  • The way the research was conducted
  • The need for research and change to practice
  • The aims of the researchers and practitioners

Our research used an online survey which was completed by 153 UK policing personnel, including 32 who work in ID suites (and whose work concentrates on obtaining eyewitness identification evidence). The respondents were generally experienced personnel, with less than 5% having less than 5 years experience, and the majority more than 16 years.

In terms of dissemination, we asked questions regarding practitioners’ knowledge of research and the recommendations that have followed from this. Overall, knowledge was generally poor, with the modal (most common) response being “I don’t know anything [about research]”. Even amongst staff working in eyewitness identification (ID) suites, only 11% indicated they tried to keep up to date with identification research, which was less than the 15% who indicated they knew nothing about this research. A similar pattern was apparent when respondents were asked about the recommendations that have been made from research, with over 50%  indicating that they didn’t know such recommendations even existed.

A number of questions were asked concerning research methodology and analyses used in the witness identification literature. A general picture emerged suggesting that one of the major barriers to putting research findings into practice may be due to the complexities of the research process itself.  For example, the responses collected seemed to indicate that participants saw research in this area as overly complex (in terms of both the methods and analyses employed), which in turn led to conclusions that were also too complex and did not speak to operational needs in a way that was implementable.

When asked about their thoughts on whether current ID procedures were effective or not, the modal response was “They generally work well and don’t need much improvement”, with over 85% of ID personnel selecting either that option or “They work very well”. Although the consensus amongst academics is that significant change is required, this is an opinion that does not seem to be shared by practitioners, and could help explain why there is little engagement with research evidence. After all, if you don’t think the system is ‘broke’, why would you look for research on how to ‘fix it’?

Perhaps the most illuminating question was to do with the type of changes that practitioners thought should be researched and implemented. The data summarised in Figure 1 show that the practitioners surveyed overwhelmingly thought that any potential changes should be aimed at increasing positive identifications – that is, a ‘hit’ outcome where the witness selects the suspect from an ID procedure. There was a difference in opinion as to whether this should not be at the cost of also increasing misidentifications (a ‘false alarm’ outcome where the witness selects the suspect from the ID procedure, but the suspect is actually innocent of the crime being investigated), with ID practitioners favouring increasing positive IDs even at the cost of misidentifications.

Changes to ID procedure

Figure 1 What should changes to ID procedures aim to achieve?

What makes this an interesting result is that it represents a fundamental clash with the aim of most research in the area, which is to reduce misidentifications, even if this was at the expense of a reduction in positive identifications – this is represented by the green bar in Figure 1. In other words, researchers have generally been concerned with finding ways of preventing miscarriages of justice, whilst ID practitioners are concerned with increasing conviction rates. To make matters worse, we know that any attempt to increase positive IDs will also result in an increase of misidentifications and likewise any attempt to reduce misidentifications will result in fewer positive IDs. Thus, the aims of researchers and practitioners are fundamentally at odds.

To conclude, this research shows that in addition to any organisational and cultural barriers to translating evidence into practice, in the critical case of eyewitness identification practitioners know very little about research findings and believe research methods, analyses and conclusions to be too complex. Perhaps more importantly, there also seems to be fundamental differences between researchers and practitioners in terms of opinions as to the current state of practice and to what the goals of research in this domain should be.

This article summarises preliminary findings that were presented as:

Harrison, V. and Pike, G. (2015).  Police perceptions of eyewitness evidence and research. Paper presented at the European Association of Psychology and Law, Nuremberg, 2015.

What can visual attention research tell us about the reliability of eyewitness evidence?

Gemma Briggs and Louise Westmarland

International Centre for Comparative Criminological Research

The Open University

Most people probably think that they would be able to provide a good account of what happened if they witnessed a serious crime. Being able to identify the person who attacked them, for example, from a line up, might seem to be a fairly straightforward matter. However, a recent report from a committee tasked with assessing the factors affecting eyewitness testimony, recommended that all law enforcement agencies should ‘…provide their [personnel] with training on vision and memory and the variables that affect them’. Whilst it may appear to be common sense that research-based policing practice should be the norm, it is evident that this is not always the case. Why, then, has this recommendation been made and how could such training help to reduce erroneous convictions?

Fredrik Fasting Torgersen

Fredrik Fasting Torgersen in the centre of a police lineup

Initiatives such as the Innocence Project highlight an alarmingly high number of cases where individuals have been convicted of crimes they did not commit on the basis of eyewitness identification. Whilst advances in forensic techniques, such as DNA testing, can help overturn some such miscarriages of justice, where forensic evidence isn’t available there still remains a tendency for jurors to assume that a confident eyewitness provides reliable evidence (Smith et al, 1989). So how should eyewitness evidence be regarded by jurors? Many researchers would claim that such evidence should be regarded with caution. This is because although police interviewing strategies, used in the UK at least, can help to focus a witness’s account of an event before their evidence is presented in court, the reliance on the witness’s memory remains. An eyewitness to a crime may feel very confident that their recollection of the event represents clearly what happened, regardless of the speed or traumatic nature of the event. They may also feel highly motivated to help the police secure a prosecution.

The desire of a witness to help the police and the confidence with which they present their account of events represent different problems for assessing eyewitness evidence. The police can control so called system variables – the methods and procedures employed to gather information from a witness, such as the cognitive interview and identification line up procedures – yet they can’t control factors such as the impact of stress, anxiety, distance from the perpetrator or the visual conditions at the time of witnessing the crime – the estimator variables (Wells, 1978). This type of ‘noise’ can affect how eyewitnesses process the crime as well as how they later recall the events. For example, an individual who witnesses a street robbery in a dark alley as they walk home from work may experience anxiety at being physically close to the attacker and fear for their own safety when the assailant sees them phoning the police. In these conditions, they may struggle to provide a clear description of the attacker or the sequence of events. Whilst poor lighting could explain the inability to describe the perpetrator’s face, the effect of fear, anxiety and distraction may be less obvious. Afterwards, the witness may feel able to provide an accurate description of the sequence of events and the physique of the attacker, which could potentially help police with their investigation, yet the emotion and distraction they experienced may well have altered their perception of the scene (Easterbrook, 1959). All of these factors, which are outside of the police’s control, are estimator variables.

scared eye

A great deal of research has been conducted on memory for, and recall of, witnessed crimes, demonstrating the constructive nature of memory (Garry et al., 1996), the effect of leading questions (Kassin and Gudjonsson, 2004) and co-witness effects on accounts (Gabbert et al., 2003), as well as the suggestibility of vulnerable witnesses (Henry and Gudjonsson, 2007). This type of research has proven beneficial in controlling some of the system variables which could affect the evidence provided by eye witnesses. However, estimator variables are harder to pin down. Research on visual attention helps to highlight how what we perceive may differ from what is actually presented to us, and that simply looking at a scene doesn’t necessarily equate to seeing all aspects of that scene (Mack and Rock, 1997). The phenomena of change blindness (the inability to detect changes in a scene despite full attention being paid); and inattentional blindness (the inability to detect changes in a scene due to attention being allocated elsewhere, or being overloaded) can both demonstrate how a witness’s perception of an event may be affected, potentially leading to incorrect identifications.

Classic change blindness investigations (e.g. Simons and Levin, 1998) have shown that people can fail to notice a change in the identity of a person they are talking to if their contact with that person is briefly interrupted (e.g. a person ducks behind a counter and a different person then stands back up). If people fail to notice a change in the identity of someone they are talking to, how good are they at noticing changes from the perspective of an onlooker?  Fitzgerald et al. (2014) showed participants a film of an innocent man entering a building. A theft then occurred, and a different (guilty) man left the building. 64% of participants who viewed the film failed to detect the change in identity. They were then shown a 6 person line up which was either target absent (contained the innocent man and 5 fillers) or target present (contained the guilty man and 5 fillers) and asked to rate their confidence in their identification.  The 36% of participants who detected the change reported higher confidence in their identification accuracy than those who demonstrated change blindness, despite their identification accuracy not being significantly greater: those who demonstrated change blindness were more likely to misidentify an innocent filler in the line up (44%) than those who had detected the change (21%), but both groups showed similar rates of correct identifications in the target present line up

In a similar study, Smart et al. (2014) questioned whether expertise could affect change blindness susceptibility and identification accuracy. They showed groups of students and law enforcement officers a video of a staged traffic event, involving a speeding car being stopped by a police officer, during which the identity and clothing of the driver changed. Participants were asked whether they detected a change in identity, a change in clothes or both before viewing a line up containing six people. There were four different versions of the line up available: two where the target was absent (i.e. only innocent fillers were present), one containing driver 1 and another containing driver 2). Results revealed that students and law enforcement officers showed similar levels of change blindness (37.5% and 50.8%, respectively) but that students were significantly more likely to detect both the change in identity and the change in clothing than police officers (22.5% and 1.6% respectively). Further, only 3.3% of police officers detected the change in clothes, compared to 40% of students. Moreover, whilst those who detected the change in identity made more correct identifications than those who did not, police officers were less accurate on the identification task than students and demonstrated a greater likelihood of selecting an innocent person from a target absent line up. Smart et al. claim that their findings demonstrate that everyone is prone to change blindness and the ‘attentional set’ an individual applies to a scene may affect detection of different changes to that scene (e.g. officers were attending more to the driver’s behaviour than his clothing).

If the attentional set, goals and level of arousal of a viewer can affect what they see when witnessing a crime, it is easy to understand how two witnesses can provide different accounts of the same event. When witnessing a crime, an eyewitness isn’t just a passive surveyor of the events – they bring with them their own goals, past experiences and current concerns – all of which can affect what they look at, process, perceive and remember of the event. Given these factors, along with research into memory and identifications, it is crucially important that eyewitness accounts are treated with caution both by the police and jurors at trial. Educating and informing these groups of the potential pitfalls of eyewitness testimony at both the time of witnessing a crime and at police interview can only be a positive thing.

Harmful evidence – wrongful conviction or suspicion on the basis of flawed eyewitness testimony

Helen Kaye, Deborah H. Drake & Graham Pike

International Centre for Comparative Criminological Research

The Open University

October 2014

You’ve probably heard people say “I never forget a face” and “I’d definitely recognise him again if I saw him”. But is this really the case, and how should an eyewitness’s memory of the face of the perpetrator be used within the criminal justice system?

Although warnings are given to juries regarding the potential unreliability of eyewitness testimony, not least because research has indicated that our memory of “who done it” may not always be as good as we think, such evidence still features very prominently in most criminal investigations and trials. In addition, eyewitness testimony is often the only direct evidence available that a particular person committed a particular act. Although forensic evidence, such as fingerprints and DNA analysis, can be very useful in establishing that a person was at a particular place or handled a particular object, it only provides indirect evidence that they committed the crime. For example, you may well be familiar with the potential problem of using indirect evidence to determine what happened from popular crime dramas, which often feature the situation where an innocent person helps a victim, and in so doing puts their fingerprints on the murder weapon and gets the victim’s blood on their clothes.

In 1992, the Innocence Project was established in the USA. This project takes up the cases of people who claim to have been wrongly convicted of a crime where DNA testing could potentially exonerate them. To date, 318 convictions have been overturned and innocent victims of injustice released, included eighteen men who were on death row. The average period of wrongful imprisonment is nearly fourteen years. The Innocence Project website reports that “Eyewitness misidentification is the single greatest cause of wrongful convictions nationwide, playing a role in 72% of convictions overturned through DNA testing”. The problem of erroneous eyewitness identification is not confined to the USA: in 2009 William Mills received a nine year jail sentence after being convicted of robbing a Glasgow bank. A critical part of the evidence was the identification by two police officers of Mills from CCTV footage. Two customers who had been in the bank at the time also picked Mills out from an identity parade. Mills was cleared of all charges and released six months into his sentence after DNA found at the scene was subsequently matched to a different person.

In 2005, the first Innocence Project was set up in the UK at The University of Bristol Law School. Although the achievements of Innocence Projects and other similar organisations are important, and indeed life changing for many, they can only correct miscarriages of justice in cases where alternative evidence is present. DNA left at the scene of the crime is often used; however, frequently the perpetrator avoids leaving samples, or by the time a case is reopened any traces have deteriorated. It’s therefore important that we explore why the misidentifications occur in the first place and how they might be avoided.

A first stage is to tackle the myth that we are good at identifying faces.  We are – but only in some circumstances, notably where the face is very familiar to us. However, evidence from psychology experiments shows that we are poor at recognising a face we’ve only encountered briefly. A typical experiment involves showing participants a video of a staged crime involving an unfamiliar person acting the role of perpetrator. Later the participants are presented with a set of photographs or video clips which may or may not include the perpetrator. The rate of correct identifications made varies depending on the exact procedure – but less than 50% accuracy is not unusual. Perhaps more surprisingly, Megreya and Burton (2008) demonstrated that even when people are shown a photograph of a face (the target) then immediately presented with a set of photographs and asked to identify the target, they are only accurate about 70% of the time.

The low accuracy rates reflect the fact that remembering is not like taking a photograph: we don’t record an exact copy of what we see, and what we recognise is influenced by many other factors than simply what we observed. A rather dramatic example of how recognition can be affected by extraneous factors is the case of psychologist Donald Thompson. Thompson took part in a live discussion on Australian TV and on his way home was arrested and placed in an identity parade. He was picked out by a rather distraught woman as the man who had raped her earlier that evening. Fortunately for Thompson the time of the rape coincided with his appearance on television, where, ironically, he had been debating the fallibility of eyewitness memory. Further investigation revealed that the woman was raped while she watched the TV show and she had confused her memory of the rapist with that of Thompson.

There may be little we can do about people’s poor memory, but we can at least try to ensure that when an eyewitness cannot correctly recognise the perpetrator of the crime they saw, they don’t pick out someone who is innocent. We need to understand the factors that cause people to identify someone wrongly, rather than reporting that they do not recognise the perpetrator in the identity parade. To some extent, variations in the line-up procedure address this – simply reminding the witness that the perpetrator ‘may not be present’ helps (for example, if the police had arrested the wrong person then the real perpetrator would not be in the line-up). Similarly, if the members of the identity parade are seen one at a time rather than all being present simultaneously then false identifications are reduced. It’s likely that this is because a simultaneous line-up encourages identification of the individual that best matches the memory of the perpetrator (i.e. the one that ‘looks most like’ the perpetrator). In the same way, unless he or she is specifically reminded otherwise, the witness may believe that the perpetrator is definitely present and pick the best match.

Characteristics of the eyewitness themselves also have an effect. For example, in laboratory tests children are more likely than young adults to identify wrongly a member of the identification parade as being the perpetrator. Havard and Memon (2013) devised an ingenious procedure to investigate this:  as well as the normal line up structure, an additional “person” – actually a silhouette called the “mystery man” – was included and the children were told that if they didn’t see the man who committed the crime, they were to pick the mystery man. This manipulation resulted in a massive drop in false identifications, from 75% to 40% in line-ups where the perpetrator was not present. Moreover there was no reduction in correct identifications when the perpetrator was present. However, 40% is still a worryingly high false identification rate and in a real situation could contribute to a miscarriage of justice.  Older eyewitnesses also tend to make a high number of false identifications in lab tests, which is perhaps even more of a concern. If they also do so in real situations where they may be seen as reliable trustworthy witnesses then their evidence, albeit given in good faith, may result in harm to innocent parties.

How are we to reduce harm and promote justice, then? Ensuring that best practice, informed by psychological research, is followed in the structure and conduct of identity parades is vital. But so is bearing in mind the evidence that memory is more fragile than we might imagine or hope, especially for unfamiliar faces.   Mistaken identifications that lead to wrongful convictions can be as unjust as not solving the crime at all.

When witnesses get it wrong

Prof Graham Pike, part of the OU psychology Forensic Psychology Research Group, writes about the recent Hallam case. Eyewitness testimony is a major area of applied psychology and one that is covered in the modules you’ll take over your psychology degree.

The Innocence Project is a US based organisation that uses DNA testing to exonerate people that were wrongfully convicted of a crime. Of the 289 cases that they have successfully overturned, more than three-quarters involved misidentification by an eyewitness. For more than 30 years, research in psychology and the social sciences has documented just how unreliable eyewitness evidence can be, and demonstrated that even a witness who is completely confident can be completely wrong.

But it is not only in the US that unreliable evidence from eyewitnesses has led to wrongful convictions. In May 2012, and after having spent 7 years in prison, Sam Hallam successfully appealed his murder conviction. The original conviction was based largely on the evidence provided by two witnesses who said they had seen Hallam at the murder scene attacking the victim. Hallam’s QC (Henry Blaxland) claimed this evidence was “so manifestly unreliable that the appellant’s submission of no case to answer should have been allowed”. When interviewed by the police, Hallam provided an alibi, but this was not investigated.

As the US Innocence Project has overturned conviction after conviction, it has been tempting to see misidentification as somewhat of a distant problem. After all, it is only recently that individual states in the US have begun to provide guidelines concerning the construction and conduct of identification procedures, whilst here in the UK we’ve had the Police and Criminal Evidence Act since 1984! The Hallam case really brings home just how dangerous eyewitness evidence can be and just how careful our criminal justice system needs to be when dealing with it. Even though we have carefully constructed guidelines that were informed by social sciences research (download this PDF for more info), it is still possible for terrible miscarriages of justice to take place.