NOTE: This article is taken from Forensic Psychology, pp. 160-169
Eyewitness testimony plays an important role within the criminal justice system and has, over the past four decades, emerged as a significant research area for psychologists and other social scientists. This chapter aims to provide a comprehensive overview of the key findings of an extensive literature on eyewitness identification performance, signposting both classic studies and emergent research strands. Taking the reader through the witnessing experience, from the initial encoding of the perpetrator to the final stage of delivering testimony in court, this chapter identifies the factors likely to lead to mistaken identifications. Theoretical implications and methodological difficulties associated with eyewitness research are also considered. In the second half of the chapter, the difficulties associated with identifications from closed-circuit television (CCTV) are examined and a full overview of the current UK guidelines for the conduct of identifications is provided.
Information obtained from eyewitnesses plays an important role in many forensic investigations. For instance, the positive identification of a suspect can provide major advances in an investigation (Coupe & Griffiths, 1996; Kebbell & Milne, 1998). Eyewitness testimony is also extremely influential in the courtroom where ‘few kinds of evidence are as compelling, or as damning, as eyewitness testimony’ (Overbeck, 2005, p.1895). Yet identifications are often disputed – and inaccurate. A review of DNA exoneration cases suggests that eyewitness errors have played some part in over 75 per cent of the convictions overturned by DNA testing in the United States (Innocence Project, 2007; see Scheck et al., 2000). That erroneous eyewitness testimony is a leading cause of wrongful convictions suggests that jurors fail to take into account factors which may have influenced or biased the eyewitness and led to a mistaken identification (Boyce et al., 2007; Huff et al., 1996).
In this chapter, we examine the performance of eyewitness and consider some of the key factors underpinning mistaken identifications.
Eyewitness Identification Performance: Experimental Research and the Real World
The scientific study of eyewitness identification, which emerged in a programmatic fashion during the 1970s, has mainly been conducted by cognitive or social psychologists and typically adopts a standard scientific experimental model. In the mock witness paradigm, volunteers and/or unsuspecting members of the public are exposed to a selected target (perpetrator) as part of a staged event (or simulated crime) and become eyewitnesses. As the events and target individuals are stipulated by the researcher, the nature of witness errors can be documented and systematic manipulations can be made to establish which recall and recognition errors are most likely under particular, forensically relevant conditions. Thus, the primary purpose of such experiments has been to establish cause-effect relations among variables (Wells & Quinlivan, 2008).
An important question for applied researchers and the legal fraternity concerns the extent to which the findings obtained in laboratory research can be generalised to the experience of actual witnesses. There are, of course, a number of important differences between the experience of (some) witnesses and unsuspecting participants in research. For instance, witnesses to ‘real’ crimes rarely receive a warning – or may not even be aware they have witnessed something important until after the event. A further concern frequently expressed by those in the legal system is that many of the ‘witnesses’ in such experiments are drawn from rather homogeneous samples of college students. In fact, many studies of eyewitness memory have included community-based samples (e.g. Gabbert et al., 2009; Lindsay et al., in press) while a significant body of research has examined the identification performance of different age groups, including young children (e.g. Pozzulo & Dempsey, 2006) and the elderly (e.g. Dodson & Krueger, 2006). Importantly, research consistently demonstrates that college-age students outperform other age populations. Thus, as noted by Wells and Quinlivan (2008), college-age participants may in fact underestimate the magnitude of eyewitness fallibility.
Witnesses to ‘real’ crime events may experience a higher level of emotional arousal, particularly if the witnessed incident involves weapons or violence and the witnesses feel threatened. For sound ethical reasons, researchers are typically not permitted to induce stress in experimental participants and are, therefore, unable to replicate violent crime scenarios in any meaningful way. Of course, it is also worth noting that the nature of the stress evoked in a controlled experimental setting may differ qualitatively from the stress associated with involvement in an actual criminal event. In brief, the effects of stress and enhanced emotion on memory are complicated, but the results of research conducted in ecologically sound settings suggest that memory is more likely to be impaired than enhanced in a stressful or arousing situation (e.g. Morgan et al., 2004; Valentine & Mesout, 2008).
Perhaps foremost on the minds of those reluctant to embrace scientific findings on eyewitness performance is the fact that the consequences of an identification decision diverge significantly when that decision is made in a laboratory as opposed to a police identification suite. It is difficult to demonstrate whether or not legal consequences have any actual bearing on witness identification accuracy. However, archival studies of actual witnesses to serious crimes indicate that witnesses taking part in identification parades, where they are presented with a suspect and a number of innocent ‘stand-ins’, select a foil (i.e. an innocent ‘stand-in’ or filler) up to 30 per cent on average (Slater, 1994; Wright & McDaid, 1996; Wright & Skagerberg, 2007). These archival data suggest that witnesses can be highly prone to error and do not necessarily become extremely cautious when faced with a high-stake identification decision (see Memon et al., 2003b).
Many factors may affect the accuracy of an eyewitness and the research literature examining these factors is extensive (Wells & Olson, 2003). A useful distinction between these factors was introduced by Wells (1978) who differentiated between estimator variables and system variables. System variables are factors which are (or could be) under the control of the criminal justice system, specifically identification test factors such as pre-lineup instructions, lineup composition, structure and presentation method. By contrast, estimator variables are not under control of the criminal justice system and, while these are factors which can be manipulated in research (such as exposure duration, age or race of witness, or presence of a weapon), they cannot be controlled in the actual witnessed incident. Therefore, the impact of such factors on witness accuracy has to be estimated, or taken into account, in a post-hoc manner.
Working systematically through the witnessing experience, from the encoding of the original incident to eyewitness testimony in court, this chapter examines several important estimator and system factors which have been shown to impair eyewitness identification accuracy. This is not intended to be an exhaustive review of all possible factors but rather a consideration of the more well-researched witness, perpetrator and contextual factors which provide some insight into subsequent witness identification behaviour and accuracy.
The Witnessed Event
Stable witness characteristics are not, on the whole, useful predictors of identification performance. Research examining factors such as gender, race or intelligence has not revealed any particularly robust effects indicating that members of some groups are better witnesses than others. Nor has research documented any strong relationship between eyewitness accuracy and personality factors. While a small number of studies have examined certain personality characteristics such as self-monitoring (Hosch & Platz, 1984) and trait anxiety (Shapiro & Penrod, 1986), ‘no strong theory relating personality to eyewitness identification has emerged’ (Wells & Olson, 2003, p.281).
However, the age of the witness has been consistently associated with identification accuracy, with findings for young children mapping onto the performance of older witnesses under certain test conditions. Specifically, when the originally encoded perpetrator is present in the lineup (a culprit present lineup), both young children and the elderly do not differ significantly from young adults in their ability to correctly identify the perpetrator. However, when the perpetrator is not present in the lineup (a culprit absent lineup), both young children and elderly witnesses are more likely than young adults to make a false identification of an innocent foil (see meta-analysis by Pozzulo & Lindsay, 1998). More recent research demonstrates that older eyewitnesses (e.g. 60 to 80 years) tend to make more false identifications than younger adults in both target present and target absent lineups (Memon et al., 2003a, 2002). No unifying theory has emerged to fully account for this finding across both age groups. For instance, it appears that young children’s identification performance is hampered by a ‘choosing problem’ (Brewer et al., 2005). Keast et al. (2007) also noted a marked overconfidence in children’s judgements relating to their identification decisions which suggests that children may be poor at monitoring their own memory, a conclusion consistent with the developmental literature (Howie & Roebers, 2007). The mechanisms underlying higher false identification rates for older witnesses are less well explored. Ageing is typically associated with reduced cognitive capacity (such as a decline in attentional resources, see Craik & Byrd, 1982; Salthouse 1982) and an increased reliance on a more ‘automatic’ feeling of familiarity rather than a more effortful recollection process (Jacoby, 1999; Mandler, 1980). Thus, it seems unlikely that the explanations for difficulties experienced by younger witnesses will also apply to older witnesses.
A more malleable witness factor at the time of encoding is blood alcohol level. If the witness has been drinking and is intoxicated, both encoding and storage may be impaired (Cutler & Penrod, 1995). In terms of identification performance, Dysart et al. (2002) found that participants with high blood alcohol readings were more likely to make a false identification when faced with a culprit absent identification task. While a number of explanations have been proposed to account for these findings, such as a tendency to focus on salient cues when intoxicated (alcohol myopia hypothesis), research on the performance of intoxicated witnesses is limited due to the associated methodological and ethical difficulties.
Stable factors (such as gender or age of the culprit) have little or no impact on witness ability to correctly identify the perpetrator. However, there are a number of well-documented factors that can serve to either impair or enhance recognition ability. For instance, distinctive faces are far more likely to be correctly identified than non-distinctive faces. Similarly, and perhaps due to their distinctiveness, attractive faces are also more easily identified than less attractive or more typical faces. The psychological mechanisms underlying these findings are relatively straightforward. When an encoded face is distinctive or atypical in some way, it will not only attract more attention and greater processing resources but the distinctive feature is also more likely to benefit from an enhanced representation in memory (Ryu & Chaudhuri, 2007; see Brewer et al., 2005 for an interesting examination of the role of distinctiveness).
Unsurprisingly, disguises usually have a negative impact on identification ability (Cutler et al., 1987; but see O’Rourke et al., 1989). Simple changes, such as covering the head, wearing glasses, growing facial hair or even altering hair style slightly, can significantly impair face recognition (Narby et al., 1996; Shapiro & Penrod, 1986). Furthermore, changes in appearance over time (such as ageing, changes in weight, etc.) also have a negative impact on identification performance. In one study, Read et al. (1990) found that photographs of a target face taken after a two-year delay were less likely to be recognised than photographs taken nearing the time of original encoding.
An extensive literature has documented the identification impairment that occurs when the perpetrator is from a different race or ethnic group to the witness. Research on own-race (also known as cross-race) bias typically demonstrates that witnesses are less accurate when attempting to identify a target from another race or ethnic group than when tasked with identifying a member of their own race (see meta-analysis by Meissner & Brigham, 2001). Specifically, research documents a higher correct identification rate from target present lineups and a lower false identification rate from target absent lineups when the witness and perpetrator are from the same race. This bias has been demonstrated in both laboratory and field studies (e.g. Wright et al., 2001) and has been observed across various combinations of ethnic groups (e.g. whites identifying blacks, blacks identifying whites, etc.). Work by Chiroro and Valentine (1995) exploring a basic contact hypothesis suggested that everyday interactions with people of different races may reduce the effect – but not consistently. Other evidence suggests that the quality rather than the quantity of cross-racial interactions may be more important in reducing own-race bias (Lavrakas et al., 1976). Interestingly, a similar pattern of results has been demonstrated for gender and age such that a match between witness and target age and gender can promote recognition accuracy (e.g. Wright & Sladden, 2003; Wright & Stroud, 2002). Taken together, these findings suggest a somewhat preferential processing mechanism for familiar stimuli. In this vein, McClelland and Chappell (1998) have argued that ownrace faces may benefit from more accurate and efficient processing due to their familiarity.
In any witnessed incident, there may be a number of situational factors which impinge on subsequent eyewitness performance. An important factor which has received surprisingly little attention from researchers is the nature of the exposure duration (i.e. the opportunity, or length of time, the witness had to observe the perpetrator). In their meta-analysis of face recognition studies, Shapiro and Penrod (1986) found the predicted linear relationship between exposure duration and hit rates (i.e. as the amount of time spent viewing the target increases so does the likelihood of a correct recognition decision). Only a handful of studies have systematically manipulated exposure duration in an eyewitness context. These studies have typically demonstrated the expected beneficial effect of longer exposure duration on subsequent identification accuracy (e.g. Memon et al., 2003a; Read, 1995). However, inconsistent choosing patterns in target absent conditions require further experimental examination. Similarly, relatively little research attention has been paid to the effect of distance on identification and the ability of eyewitnesses to correctly estimate distances from an incident or perpetrator. Obviously, a correct identification is somewhat unlikely if the witness was unable to see the perpetrator so research has tended to focus on identifying a useful ‘rule of thumb’ with respect to distance. For instance, Wagenaar and van der Schrier (1996) suggested that identification performance was optimal when the viewing distance was less than 15 metres from the target. However, recent work by Lindsay et al. (2008) reveals that the 15-metre rule may not be useful – or accurate – for two reasons. Firstly, if witnesses are unable to estimate distance reliably then they are unlikely to be able to report accurately whether they were less than 15 metres from the target. Secondly, it seems rather unlikely that all identifications made when the viewing distance was less than 15 metres will be correct – or vice versa. In Lindsay et al. (2008), over 1300 participants observed a target person at various distances, estimated the distance to the target, generated a description and attempted an identification of the target from either a target present or target absent lineup. Participants were poor at accurately estimating the distance between themselves and the target (particularly when required to make this estimate from memory). While the reliability of target descriptions was unimpaired up to distances of approximately 50 metres, a decline in identification performance occurred for both target present and target absent lineups as distance between the witness and target at encoding increased. Although this finding is broadly consistent with those of Wagenaar and van der Schrier (1996), Lindsay et al. (2008) did not observe any dramatic dropoff in identification accuracy at 15 metres, noting that many participants made correct identifications beyond this distance, suggesting that a 15-metre rule is not a particularly useful diagnostic for the courts.
Another variable aspect of a criminal incident is the amount of stress or fear a witness may experience. Research inducing realistic levels of stress is, for obvious methodological and ethical reasons, difficult to conduct. However, in a field training scenario, Morgan et al. (2004) subjected soldiers to either a high- or low-stress interrogation in a mock prisoner of war camp over a 12-hour period. After a 24-hour delay, soldiers who had experienced a high-stress interrogation were significantly less likely to correctly identify their interrogator than those who had experienced the low-stress interrogation. A more recent study conducted on civilian participants in an arousing context (the London Dungeon) demonstrated that high-state anxiety was associated with fewer correct identifications of a target (Valentine & Mesout, 2008).
Other researchers have focused on the forensically relevant problem of witnesses to crimes involving weapons. Although some field research suggests that the emotional arousal associated with violent witnessing conditions may actually serve to benefit memory (e.g. Yuille & Cutshall, 1986; but see Wright, 2006), eyewitness experts have tended to favour the view that incidents involving the presence of a weapon will have a negative impact on eyewitness performance (Kassin et al., 2001). This phenomenon has become known as the weapon focus effect (Loftus et al., 1987) and occurs when the presence of a weapon adversely affects subsequent eyewitness recall performance such that memory for details such as the perpetrator’s facial characteristics and clothing is impaired (e.g. Cutler et al., 1987; Hope & Wright, 2007; Loftus et al., 1987; Maas & Kohnken, 1989; Pickel et al., 2003; Steblay, 1992). One explanation is that increased arousal (or stress) due to the presence of a weapon reduces attentional capacity as increased attention is paid to the weapon while peripheral cues are ignored or filtered (Hope & Wright, 2007; Loftus, 1980; Macleod & Mathews, 1991). A meta-analytic review of the effects of stress on eyewitness memory by Deffenbacher et al. (2004) concluded that high levels of stress impair the accuracy of eyewitness recall and identification but that the detriment depends on the response mode elicited by the stress manipulation. The authors propose that some emotion manipulations generate an ‘orientating’ response while others generate a ‘defensive’ response (Deffenbacher, 1994; Deffenbacher et al., 2004; see also Klorman et al., 1977; Sokolov, 1963). Deffenbacher et al. (2004) argue that the orientating response leads to enhanced memory for ‘informative aspects’ of a scene but that the defensive response can lead to either enhanced memory or significant memory impairment depending on other cognitive and physiological factors.
Between the Witnessed Event and Identification Task
In the delay between an individual witnessing a crime and making an identification attempt, the witness’s memory is not only prone to decay, but it is also vulnerable to the influence of post-event information from numerous sources. Both delay and post-event information have been shown to compromise recall completeness and accuracy (see Anderson, 1983; Ayers & Reder, 1998; Ellis et al., 1980; Gabbert et al., 2003; Loftus et al., 1978; McCloskey & Zaragoza, 1985; Meissner, 2002; Tuckey & Brewer, 2003).
Delay systematically decreases the amount of information that can be recalled (Ebbinghaus, 1885; Kassin et al., 2001; Rubin & Wenzel, 1996; see also Tuckey & Brewer, 2003). Furthermore, a meta-analysis of 128 studies of face recognition suggests that there is a decline in the correct identification of previously seen faces after a delay (Shapiro & Penrod, 1986). Sporer (1992) found a decrease in correct identifications and an increase in false alarms over various intervals up to three weeks. Importantly, the field work by Valentine et al. (2003b) examining performance of real witnesses suggests that the greatest decline in performance occurs when the delay exceeds one week.
Research conducted by Elizabeth Loftus in the 1970s demonstrated the misinformation effect – a powerful phenomenon resulting in memory distortion (for a review see Loftus, 2005). In a now classic experiment, Loftus and Palmer (1974) presented participants a short film of a car accident and subsequently tested participant recall for details of the incident. Importantly, they found that simply changing one word in a question pertaining to the speed the car was travelling when the accident occurred resulted in significantly different estimates of speed. Specifically, participants asked to estimate what speed the car was travelling at when it contacted the other vehicle provided slower speed estimates (31.8 mph) than those asked to estimate the speed of the car when it smashed into the other vehicle (40.5 mph). Including the verb ‘smashed’ in the question also led to increased false reports of witnessing broken glass at the scene of the accident (no broken glass was ever shown). Several hundred experiments since have demonstrated the misinformation phenomenon, explored boundary conditions of the effect and served the development of theoretical explanations. More pertinent to eyewitness identification accuracy is an emerging body of work on the impact of co-witness influence on memory. In a recent survey 86 per cent of real eyewitnesses discussed their memory with a co-witness who was present at the witnessed event (Paterson & Kemp, 2006). Witnesses to an event may share the same experience but their individual recall of the event may differ for many reasons, including naturally occurring differences in attention paid to various details of the event, differences in spatial or temporal location at the scene or perceived differences in ability to recall those details (Gabbert et al., 2006).
Research amply demonstrates that the most likely outcome when two witnesses discuss their memories is that their accounts of the witnessed event become more similar and, hence, seemingly corroborative (Gabbert et al., 2004; Wright et al., 2000). A witness is also more likely to be influenced by a co-witness with whom they have a prior acquaintance, such as a friend or partner (Hope et al., 2008). However, very few studies have explored the impact of misleading information on subsequent identifications. A recent study conducted by Gabbert et al. (2007) which manipulated co-witness confidence and accuracy across both target present and target absent lineups found that participants were more likely than controls to reject the lineup incorrectly when they were aware that the co-witness had rejected the lineup. However, participants were no more likely than controls to identify the perpetrator correctly after seeing the co-witness make an accurate identification, and the pre-lineup confidence expressed by the confederate did not appear to influence the witness.
While unbiased lineup procedures may ensure that identification decisions themselves are unlikely to be shared with other witnesses, misinformation concerning descriptive details or pertaining to the general appearance of the target may have a negative impact on eyewitness accuracy, and this hypothesis is worthy of further experimental scrutiny.
Intermediate Recognition Tasks
In the course of an investigation, witnesses may be asked to search through a set of mugshots (usually photographs of potential suspects). Unsurprisingly, a number of studies have shown that previous exposure to the suspect increases the likelihood that the suspect will be identified in a subsequent lineup. In other words, repeated exposure to a suspect can increase mistaken identifications of an innocent suspect (Brigham & Cairns, 1988; Dysart et al., 2001; Gorenstein & Ellsworth, 1980; Memon et al., 2002).
In an investigation where no suspect has emerged, the police may work with a witness to produce a facial composite of the perpetrator. Previously, this composite might have been produced by a sketch artist but technological advances have led to the use of computerised systems for composite production (such as the E-Fit or Profit identification systems). While research demonstrates that the quality of composites is often rather poor, with little likeness to the appearance of the actual perpetrator (see Wells & Hasel, 2007, for a review), a more important question concerns the extent to which generating a composite might impair identification accuracy. In two studies, Wells et al. (2005) examined whether building a face composite had a negative effect on memory for the target face. Results indicated that building a composite resulted in significantly lower identifications for the original target face (Experiment 1), while a second experiment revealed that the results might be generalised to a standard witness paradigm (Experiment 2). In light of these results, Wells et al. (2005) suggest that where multiple witnesses are available ‘it might be possible to use one witness to build a composite and save the other witnesses for any later lineup identification attempts’ (p.155).
The Identification Task
In this section, we consider several important system variables which can have a significant impact on eyewitness identification performance. These variables are ultimately under the control of the criminal justice system and, to date, research has focused on demonstrating the identification errors resulting from poor practice in the production and administration of identification tests (i.e. lineups) while delivering recommendations for improved procedures.
Often witnesses assume that the suspect apprehended by the police and presented to them in the formal setting of a lineup must have a high probability of being the actual perpetrator. In other words, witnesses assume that they would not have been invited to make an identification if there was not a good reason for the police to believe the suspect was the actual perpetrator and their role is to make a positive identification (i.e. choose someone from the array). This bias may be further exacerbated if the witnesses are presented with the task in a misleading manner (i.e. ‘Take a good look at the lineup and see if you can identify the offender’). In fact, Memon et al. (2003a) found that over 90 per cent of mock witnesses indicated that they expected the perpetrator to be present in a lineup even under unbiased conditions. Therefore, it is extremely important that witnesses are informed that the person they saw ‘may or may not be present in the lineup’. Incorrect identifications from target absent lineups are significantly lower when witnesses are given this simple cautionary instruction (see meta-analyses by Steblay, 1997; Clark, 2005).
When a suspect disputes his involvement in an incident or claims an identification error, a lineup must be conducted. Here the police face a number of challenges as there are (at least) two important dilemmas with respect to lineup composition, namely, the number of lineup members (or foils) present in addition to the suspect, and how those foils are selected. The requisite number of lineup members is typically specified in law. For instance, in the UK a lineup must contain at least eight foils, while in the US lineups containing five (or more) foils are common. However, researchers have drawn a sharp distinction between the nominal size of a lineup (i.e. the number of people appearing in the lineup) and what has been described as the functional size of a lineup (Wells et al., 1979). Functional size refers to the number of plausible lineup members. If an eyewitness describes a perpetrator as a male, in his early twenties with long, dark hair, but then views a lineup in which two of the foils have short dark hair and one other foil is in his 40s, then the functional size of the lineup is reduced by three members, as these foils will be automatically discarded by the witness as they do not match the original description provided. The purpose of the lineup is to provide a fair identification task in which the suspect does not ‘stand out’ inappropriately from the other foils. Reducing the functional size of the lineup – particularly when the suspect is not the actual perpetrator – significantly increases the chance of a false identification (Lindsay & Wells, 1980; Tredoux, 2002). Thus, the selection of appropriate foils is critical for the production of a fair lineup and has been the focus of a good deal of debate. In the UK, police are required to select foils that resemble the suspect in what might be described as a ‘match to suspect’ strategy. In other words, foils are selected on the grounds that they match the appearance of the suspect (rather than the description of the perpetrator).
This strategy is problematic as research has documented that foils who do not coincide with a witness’s prior verbal description are likely to be disregarded, resulting in a biased lineup and an increased likelihood that an innocent suspect may be mistakenly identified (e.g. Clark & Tunnicliff, 2001). Thus, a ‘match to description’ strategy (i.e. where foils are selected based on their match to descriptions of the perpetrator provided by the witness) may be preferable (Luus & Wells, 1991). However, more recent research by Darling et al. (2008) did not identify any differences in either correct or incorrect identifications as a function of these lineup composition strategies. Clearly, further research is necessary to identify specifically how alterations to the composition of a lineup affect choosing behaviour.
Ideally, lineups should take place under double-blind administration where both the witness and lineup administrator are unaware of the identity of the suspect. Where the person conducting the lineup knows which lineup member is the suspect, there is a possibility that they will unintentionally transmit this knowledge to the witness (Harris & Rosenthal, 1985), resulting in increased rates of false identification (Phillips et al., 1999). More recently, Greathouse and Kovera (2008) noted that administrators displayed more biasing behaviours (such as inviting the witness to ‘take another look’, providing overt cues as to the identity of the suspect, and exerting greater pressure on witnesses to choose) during single-blind administration procedures (i.e. when they knew the identity of the suspect) than under double-blind procedures.
Lineup procedure: Comparing absolute and relative judgements
The lineup task has probably received greater research attention than any other topic relating to eyewitness testimony. In the traditional lineup (which may involve photographs or live participants, depending on the jurisdiction), the suspect and foils are presented simultaneously. Given witnesses tendency to assume that the perpetrator will be present in the lineup, the opportunity to examine all lineup members at once can lead witnesses to compare the lineup members with each other and select the lineup member who best matches their original memory. This has been described as a relative judgement strategy (Wells, 1984; Wells & Seelau, 1995). An alternative method of lineup presentation, known as the sequential lineup, was proposed by Lindsay and Wells (1985). Unlike the traditional simultaneous lineup where all lineup members are viewed at once, in the sequential lineup method each lineup member is presented sequentially, one member at a time. The witness is required to make an absolute identification decision for each lineup member (Is this the perpetrator you saw? Yes or No) prior to seeing the next person in the lineup. In the optimal version of the lineup, the witness does not know how many faces will be presented and the lineup terminates when a choice is made, with witnesses not permitted to see any further photos, review previously presented photos or change their identification decision. This lineup method promotes an absolute identification decision as, unlike the simultaneous lineup, witnesses cannot engage in relative comparisons between lineup members but instead have to compare the face presented with their memory for the perpetrator. Many studies have demonstrated that the sequential lineup method significantly reduces false identifications (see Steblay et al., 2001 for a metaanalysis) as a consequence of promoting a more conservative response criterion than the simultaneous lineup procedure (Meissner et al., 2005). However, a number of recent studies, while typically observing the predicted improvements in the false identification rate, have also noted a reduction in correct identifications under sequential procedures (Ebbesen & Flowe, 2002; Memon & Gabbert, 2003). Interestingly, in their metaanalysis, Clark et al. (2008) noted that only biased lineups produced the sequential lineup advantage with respect to false identifications. Further research is necessary to better understand the mechanisms driving choosing behaviour in order to develop accuracypromoting lineup formats.
Witness confidence is, perhaps, the most influential cue used by juries when evaluating the credibility and reliability of eyewitness testimony (Cutler et al., 1990; Lindsay et al., 1981). However, mistaken eyewitnesses can be overconfident (Shaw & McClure, 1996; Wells & Bradfield, 1999) and eyewitness confidence can be highly malleable in the period after making an identification (Luus & Wells, 1994a, 1994b; Wells & Bradfield, 1998). For instance, Wells and Bradfield (1998) found that witnesses who were given positive feedback (e.g. ‘Good, you identified the suspect’) reported higher confidence and better viewing conditions than those who received no feedback. Conversely, witnesses given negative feedback were less confident and reported worse witnessing conditions. The effects of feedback have also been shown to occur for both target present and target absent lineups (Bradfield et al., 2002), when there are long delays between identification and feedback (Wells et al., 2002), and even extend to witness willingness to testify (Wells & Bradfield, 1998, 1999). Post-identification effects may be reduced (but not eliminated) by means of warnings (e.g. Lampinen et al., 2007).
Is confidence ever related to accuracy?
Police, lawyers, judges and other legal practitioners, in addition to lay jurors, typically consider confidence as a useful indicator of likely eyewitness accuracy (Deffenbacher & Loftus, 1982; Noon & Hollin, 1987; Potter & Brewer, 1999). As we have seen, however, eyewitness confidence is malleable and susceptible to bias which can, in the worst-case scenario, produce highly confident mistaken identifications. But can witness confidence actually tell us anything useful about identification accuracy? Until recently, researchers have tended to take the view that confidence is not reliably associated with accuracy and, in particular, is not a reliable predictor of accuracy given low or non-significant confidence– accuracy correlations (e.g. Bothwell et al., 1987; Sporer et al., 1995; see also Kassin et al., 2001). However, in an extensive programme of research focusing on confidence and adapting alternative analyses, Brewer and his colleagues have challenged this conclusion (Brewer, 2006; Brewer & Wells, 2006; Weber & Brewer, 2004). Using a calibration approach, these authors have documented substantial confidence–accuracy relations for lineup choosers (i.e. witnesses who make positive identifications) across various stimuli materials (for extended discussion of this method and the relationship between confidence and accuracy, see Brewer, 2006; Brewer et al., 2005).
Identifications from CCTV
Intuitively, one might expect that identification performance might improve significantly when the ‘witness’, be that the original witness, a CCTV operator or police officer reviewing the evidence, has access to a video recording of the (alleged) target and, possibly, still photographs of the suspect. With video footage of the incident available, the task would no longer rely so heavily on memory (or prior familiarity with the perpetrator) and would simply require the witness to engage in an apparently simple matching task. However, the identification of individuals from CCTV footage is not necessarily a simple identification task and, like other identification tasks, is prone to error – even under optimal conditions.
There are two quite distinct circumstances where an attempt may be made to identify a face from a video image (Bruce et al., 1999). In the first situation, a spontaneous identification may be made by a member of the public (or perhaps, a CCTV operator or police officer) who claims that the target appearing in the CCTV image is personally known to them. In the second situation, the target appearing in the CCTV footage is compared to an apprehended suspect to establish whether, in fact, the suspect was recorded at the scene of the incident under investigation. Identification accuracy varies under these circumstances with respect to whether the face is previously known or previously unknown to the witness.
In one of the early studies on spontaneous identifications based on prior exposure, Logie et al. (1987) examined the ability of the general public to identify a live target in a town centre from a previously presented photograph. The photograph had been published in a local newspaper. Despite circulating details of the precise location of the target, the spontaneous detection (i.e. identification) rate for the general public was very low and this was coupled with a high false recognition rate (i.e. false identifications of other ‘innocent’ passers-by).
These low recognition rates in dynamic interactions where the target face is continually available to the witness have been documented elsewhere. In a field study, Kemp et al. (1997) examined whether credit cards bearing a photograph of the cardholder might serve to reduce credit card fraud. Including a photograph of the legal cardholder on a credit card (or indeed, other identity document) would seem to be a relatively foolproof method of ensuring that the card is used only by the person entitled to use it. In their study, shoppers presented a credit card bearing a photograph of themselves to pay for half the transactions while for other transactions they presented a card bearing the photograph of another individual. Experienced checkout cashiers were required to either accept or decline the card depending on their verification of the cardholder’s identity, and rate their confidence that the photograph appearing on the card was, in fact, that of the shopper. More than 50 per cent of the fraudulent cards were accepted by the cashiers – despite the fact that cashiers were aware that a study was under way and indicated that they both spent longer examining cards and had been more cautious than usual.
High error rates in the ability to match a target from CCTV footage have also been documented. Typically, it has been assumed that difficulties in identifying faces from video recordings are largely due to the frequently poor-quality nature of the recording and that were high-quality recordings available such difficulties would be reduced. While it is true that many CCTV images may be of poor quality for a number of technical reasons (such as unsuitable lighting conditions, intermittent image sampling, etc.), the assumption that this alone underpins low accuracy rates in face matching from CCTV has been challenged by research findings.
Bruce and her colleagues (1999) examined how well people were able to match faces extracted from a high quality video-recording against high-quality photographic images. The results revealed that overall accuracy was relatively poor (averaging only 70 per cent across trials) even under these optimal conditions. Performance was further degraded when the target expression or viewpoint was altered. Furthermore, the use of colour target images (as opposed to black-andwhite images) did not appear to lend any particular advantage (or disadvantage) to performance on the matching task. Thus, it would appear that our ability to identify an unfamiliar face – even in the presence of a reference image (such as a CCTV still or a photograph) is surprisingly error-prone (Davies & Thasen, 2000; Henderson et al., 2001).
In contrast, identification accuracy for known or familiar faces can be very accurate – even when the target images are of poor quality. To examine the impact of familiarity on face recognition, Burton et al. (1999) showed study participants surveillance video footage of a target who was known to some participants but not others. Results indicated a marked advantage for people who were personally familiar with the target – 73 per cent of the poor-quality image targets were recognised when they were familiar. In a series of studies exploring the role of familiarity, Bruce et al. (2001) found that participants were able to correctly verify (or reject) a familiar target with a high degree of accuracy (over 90 per cent) despite the use of poor-quality video images. When participants were unfamiliar with the targets, the accuracy rate was significantly lower (56 per cent). Subsequent experiments revealed that brief periods of exposure to the target do not necessarily generate sufficient familiarity to improve the recognition or matching of unfamiliar faces – unless some ‘deep’ or social processing has taken place (i.e. discussing the faces with another person).
Face recognition is of central importance to investigative police work (Scott-Brown & Cronin, 2007). CCTV has the benefit of providing investigators with a permanent record of an event and, importantly, who may have been involved in it. The availability of CCTV footage – and the speed at which it was analysed – facilitated the rapid identification of the 7/7 and 21/7 bombers from thousands of hours of recordings (Metropolitan Police, 2005). Furthermore, actual CCTV footage is generally considered powerful evidence in court (NACRO, 2002; Scott-Brown & Cronin, 2007; Thomas, 1993). However, relying on CCTV for the recognition and identification of suspects may foster a false sense of security and a potentially dangerous over-reliance on such evidence. We expect to be able to do this task with a high degree of accuracy. However, the research consistently demonstrates that people are poor at this task – even under optimal conditions.
Is eyewitness identification evidence reliable?
Experimental psychological research on eyewitness identification has flourished over the past 30 years, producing hundreds of articles and thousands of identification data-points. Given the size of the literature and the many different designs and research hypotheses deployed, it is often difficult to compare between studies and reach an overall conclusion with respect to our ability to identify correctly a previously seen individual. As Clark et al. (2008) note, correct identification rates often vary widely across experiments, for instance from as high as 80 per cent to as low as 8 per cent. To establish what the results of eyewitness experiments can tell us, Clark et al. (2008) conducted a meta-analysis of 94 comparisons between target present and target absent lineups. The most important conclusions to emerge from this analysis were as follows: 1. correct identifications (from target present lineups) and correct non-identifications (target absent lineups) were not correlated; 2. an identification of the suspect is diagnostic of the suspect’s guilt but the identification may be less informative if any of the identification procedures are in any way biased (such as lineup composition); and 3. non-identifications were diagnostic of the suspect’s innocence while ‘don’t know’ responses were, unsurprisingly, non-diagnostic with respect to guilt or innocence. Based on these and earlier analyses (e.g. Clark, 2005), Clark et al. (2008) suggest as a basic principle that ‘a suspect identification has greater probative value to the extent that it is based on the witness’s memory, and less probative value to the extent that it is due to lineup composition or an increase in the witness’s conformity, willingness, or desire to make an identification’ (p.211). Thus, when assessing the reliability and likely accuracy of an identification, legal practitioners and juries alike need to consider the extent to which these factors might have played a role in the identification process.
The Eyewitness in Court
The final stage of the eyewitness’s role within the legal process takes place in court. Courts in many jurisdictions acknowledge that there is a risk that eyewitness evidence may be unreliable and jurors are typically instructed to scrutinise the circumstances under which the witness encountered the suspect (Memon, 2008). For instance, in England and Wales trial judges are required to ‘protect against unsafe convictions in cases involving disputed identification’ (Roberts & Ormerod, 2008, p.74). The Turnbull guidelines (R v. Turnbull) stipulate that if a prosecution case is heavily based on eyewitness identification evidence, where the judge considers that evidence to be weak, of poor or questionable quality, the case must not proceed. When a case involving eyewitness identification evidence does proceed before a jury, the judge is required to provide both a general warning regarding the risks associated with eyewitness evidence and a more specific warning tailored to the nature of the potential weaknesses of the eyewitness evidence in that particular case.
The admissibility of expert testimony concerning eyewitness testimony remains a topic for some debate in legal circles (see Benton et al., 2006 for a review), In most adversarial systems, including North America and the UK, the judge decides whether expert testimony is admissible against certain criteria (Benton et al., 2006; Kovera et al., 2002; Read & Desmarais, in press). The one criterion common across most jurisdictions concerns the extent to which issues pertaining to eyewitness memory are considered to be a matter of juror common sense. In the UK this means that the jurors are usually expected to make a sound decision about the quality of eyewitness evidence unaided by testimony from an expert. The judicial conclusion that eyewitness memory is indeed a matter of common sense is one of the most frequently cited reasons for the rejection of eyewitness expert testimony (Benton et al., 2006; Leippe, 1995; Yarmey, 2001), and legal experts are often in agreement (e.g. Benton et al., 2006; Stuesser, 2005).
However, jurors are not particularly sensitive to potential eyewitness error – or responsive to judicial instructions on the matter (Kassin & Sommers, 1997). In fact, over a quarter of a century of research has demonstrated that lay understanding of eyewitness psychology is limited – and often mistaken (e.g. Benton et al., 2006; Brigham & WolfsKeil, 1983; Deffenbacher & Loftus, 1982; McConkey & Roche, 1989; Noon & Hollin, 1987; for a comprehensive review see Benton et al., 2006). Jurors tend to be unaware of the implications of biased procedures used by law enforcement, such as poorly constructed lineups, misleading feedback or biased instructions (Shaw et al., 1999). Potential jurors also find it difficult to distinguish between accurate and inaccurate witnesses (e.g. Lindsay et al., 1989, 1981). Even legal professionals are typically rather limited in their understanding of factors affecting eyewitness accuracy (Granhag et al., 2005; Wise & Safer, 2004). Furthermore, convictions which originally relied heavily on eyewitness testimony, but are now known to have been in error, illustrate quite clearly that jurors are often unable to either generate or apply the common sense expected of them by the courts.
Eyewitnesses serve an important function in the delivery of justice and can, under the right circumstances, correctly confirm the identity of a criminal. However, caution needs to be exercised with respect to identifications as the leading cause of mistaken convictions is erroneous eyewitness testimony. In particular, consideration must be paid to the conditions under which the witness encoded the perpetrator, the presence of any intervening misleading information, the nature and fairness of the identification procedures and whether the witness received feedback – unwittingly or otherwise.
Brewer, N., Weber, N. & Semmler, C. (2005). Eyewitness identification. In N. Brewer & K.D. Williams (Eds.) Psychology and law: An empirical perspective (pp.177–221). New York: Guilford Press.
An excellent and thoughtful overview of key issues in eyewitness research, Specifically, this chapter examines the various stages of the identification process that occur in the real world, from features of the event which may impede the witness to the impact of exposure to inaccurate post-event information and, finally, the identification task. Brewer and his colleagues also critically examine other factors which research suggests may be diagnostic of identification accuracy (e.g. confidence and latency). Throughout the chapter, the authors highlight several important methodological shortcomings which beset the extant research literature, such as underpowered experiments, a limited stimulus set and inadequate lineup conditions. Not only does this chapter provide a comprehensive review of the eyewitness literature and consider some of the problematic methodological issues faced by researchers but, importantly, it also focuses on the need to further develop our theoretical understanding of eyewitness identification behaviour.
Valentine, T. & Heaton, P. (1999). An evaluation of the fairness of police lineups and video identifications. Applied Cognitive Psychology, 13, S59–S72.
Valentine’s work evaluating the fairness of VIPER lineups makes an important contribution to our understanding of current UK identification procedures. In this initial study of video identifications, Valentine and Heaton compared the ‘fairness’ (in terms of non-biased lineup selections) of photo versus video identification stimuli. In a fair lineup the suspect should be chosen, by chance, by 11 per cent of the mock witnesses (i.e. each lineup member should have an equal chance of being selected if the actual perpetrator is not present and correctly identified). However, in this study, 25 per cent of mock witnesses selected the suspect from the photographs of live lineups while only 15 per cent of mock witnesses selected the suspect from video lineups. The authors concluded that the video lineups were fairer than the live lineups. Given that mistaken eyewitness identifications are a significant source of miscarriages of justice, Valentine and Heaton argue that the more widespread use of video identification may actually improve the reliability of identification evidence.
Weber, N. & Brewer, N. (2004). Confidence–accuracy calibration in absolute and relative face recognition judgments. Journal of Experimental Psychology: Applied, 10, 156–172.
This paper introduces an important new conceptual and analytical approach to eyewitness confidence which continues to show promise in determining the likely diagnosticity of eyewitness identification decisions. Confidence–accuracy calibration was analysed for both absolute and relative face recognition judgements. The most interesting finding is that recognition judgements for ‘old’ (i.e. previously viewed) stimuli demonstrated a strong confidence–accuracy calibration. In other words, there was an association between accuracy and the level of confidence expressed. This finding suggests that there was a meaningful relationship between subjective and objective probabilities of judgement accuracy for previously seen items. However, for ‘new’ judgements there was little or no association between confidence and accuracy using the calibration approach. See also: Brewer, N. (2006). Uses and abuses of eyewitness identification confidence. Legal and Criminological Psychology, 11, 3–23.
Wells, G.L., Memon, A. & Penrod, S. (2006). Eyewitness evidence. Improving its probative value. Psychological Science in the Public Interest, 7, 45–75.
A thorough review of the eyewitness literature and its role within the legal system. In this article, both estimator and system variables are examined and, in particular, the authors focus on how procedures based on scientific research findings can be developed to improve the probative value of eyewitness evidence. Other important questions are addressed, including the frequently occurring tension in applied research between scientific rigour and external validity when moving from the laboratory to real-world contexts. Specifically, the authors consider issues of base rates, multi-collinearity, selection effects, subject populations and psychological realism and note how a combination of critical theory and field data can work together to improve the generalisability of eyewitness research.