Forensic Psychology
Page 71
When constructing a culprit-description lineup it may be necessary to take account of default values in descriptions. Sometimes people may not describe the sex or ethnicity of the person, or may neglect to say that somebody did not have a beard or was not wearing glasses. This may occur because the witness assumes a default value (Lindsay, Martin, & Webber, 1994).
Witnesses to a staged crime made significantly more correct identifications from culprit-description lineups than from suspect-resemblance lineups when the culprit was in the lineup, but the number of mistaken identifications from culprit-absent lineups did not differ reliably (Juslin, Olsson and Winman, 1996; Wells, Rydell, and Seelau, 1993). Lindsay et al. (1994), Tunnicliffe and Clark (2000) and Darling, Valentine and Memon (2008) did not find a statistically significant difference in the rate of correct or mistaken identification between culprit-description and suspect-resemblance lineups. At present, the empirical evidence is too ambiguous to recommend that a suspect-resemblance strategy should be changed to a match-to-description strategy. However, it is clearly good practice to consider whether the suspect stands out in comparison to the witness’s description. Clark (2012) found that using more similar foils in a lineup led to fewer mistaken identifications from culprit-absent lineups, but also fewer correct identifications of the perpetrator when present. However, the former effect was stronger than the latter, so eyewitness identifications were of greater probative value when the lineup contained more similar foils. A meta-analysis of the literature by Fitzgerald, Price, Oriet and Charman (2013) drew a similar conclusion.
15.6.7 Relative and Absolute Judgements: Sequential and Simultaneous Presentation
A persistent problem in understanding eyewitness identification is to explain why a sizable minority of witnesses (about one in five) make mistaken identifications, despite appropriate warnings that the perpetrator may not be present in the lineup. Wells (1993) demonstrated that at least part of the problem may be attributable to witnesses who make a relative judgement rather than an absolute judgement. When confronted with a lineup, a witness may only identify a person if their resemblance to the culprit exceeds some criterion of similarity (an absolute judgement). Alternatively a witness may examine all the members of a lineup and identify the person who most closely resembles the perpetrator (a relative judgement). The influence of relative judgements was demonstrated using a method of removal without replacement (Table 15.1). The top line of Table 15.1 shows the distribution of selections made by 100 witnesses to a mock crime, who saw a six-person lineup in which the perpetrator was present. Fifty-four percent identified the perpetrator and 21% rejected the lineup. Other participants were presented with a five-person lineup from which the perpetrator was removed. One might expect that the witnesses who would correctly identify the perpetrator if he had been present, would reject the lineup. Instead the number of mistaken identifications increases dramatically. In the absence of the culprit many witnesses appeared to identify the foil who most closely resembled their memory of the perpetrator. Clark and Davey (2005) replicated this shift of choices from the perpetrator to the foils. These data suggest that relative identification decisions may be a cause of mistaken identifications.
Table 15.1 The distribution of identifications across members of a photograph lineup with the perpetrator present and with the perpetrator removed. Data from Wells (1993)
Line-up member
1 2 3(p) 4 5 6 No choice
Perpetrator present 3% 13% 54% 3% 3% 3% 21%
Perpetrator removed 6% 38% – 12% 7% 5% 32%
The method of sequential lineup presentation was developed to impede witnesses making a relative judgment. In a sequential presentation, photographs of faces are presented one at a time (Lindsay & Wells, 1985). The lineup administrator should be blind to the identity of the suspect. The witness is not told how many faces will be presented, but must decide when each face is presented whether it is or is not the culprit before the next face is presented (Lindsay, Lea, & Fulford, 1991). Furthermore, witnesses must not be allowed a second choice or to see again a face previously presented.
Steblay, Dysart, and Wells (2011) carried out a meta-analytic comparison of the accuracy rates in sequential and simultaneous lineups. The most relevant analysis was of 27 published tests that compared sequential and simultaneous presentation of both culprit-present and culprit-absent lineups. When present in the lineup, more witnesses identified the culprit from simultaneous lineups than from sequential lineups (52% vs. 44% respectively). When the culprit was not in the lineup, there were more mistaken identifications of the designated innocent suspect from simultaneous than from sequential lineups (28% vs. 15%). Steblay et al. concluded that sequential presentation lineups produced evidence that is more diagnostic of guilt. Although meta-analytic comparisons of simultaneous and sequential procedures by other authors found broadly similar results, they drew different implications for policy (Clark, 2012; Palmer & Brewer, 2012). These differing conclusions reflect the difficulty, using conventional methods of evaluating lineup performance, in evaluating factors that have beneficial effects for facilitating identification of a guilty perpetrator, but an increased risk of mistaken identification of an innocent suspect or vice versa. Recently, researchers have advocated use of signal detection measures to measure lineup performance. Box 15.4 provides more details.
BOX 15.4 SIGNAL DETECTION THEORY AND EYEWITNESS IDENTIFICATION
If an experimental manipulation, such as sequential instead of simultaneous lineup presentation, reduced the proportion of innocent suspects identified, but increased the proportion of guilty culprits identified, which procedure is most effective? Many authors have argued that the probative value of a procedure can be measured by the diagnosticity ratio (“hit” rate from culprit present lineups/false alarm rate for innocent suspects from culprit absent lineups). However this measure is affected by the level of bias or response criterion – the propensity of witnesses to make an identification (Mickes, Flowe and Wixted, 2012; Clark, 2012).
Signal detection theory (SDT) allows sensitivity – the ability to distinguish the perpetrator from foils – to be measured independently of response criterion – the propensity to make identification. SDT has been used for many years in laboratory experiments on human memory. It was first used to evaluate lineups by Meissner, Tredoux, Parker, and Maclin (2005). From a laboratory experiment, in which participants attempted to identify several target faces from lineups, Meissner et al. concluded that sequential presentation of lineups induced a more conservative response criterion but did not affect discrimination accuracy.
Mickes et al. (2012) introduced a method for using signal detection theory for applied studies of eyewitness memory in which only one identification decision is collected from each “witness”. Data were combined from different witnesses and plotted as a function of the level of confidence to produce a receiver operating characteristic (or ROC). Mickes et al. calculated the sensitivity for simultaneous and sequential lineups. They concluded that simultaneous presentation was the most sensitive measure and therefore provided the more reliable evidence of guilt and innocence. Mickes, Moreland, Clark and Wixted (2014) describe a method to calculate sensitivity (d’ – known as d prime) when confidence data is not available. These authors argue that d’ should be used in lineup research rather than the diagnosticity ratio. A measure of performance based on SDT is now generally required to publish experiments on eyewitness identification.
Video identification used in the UK naturally yields a sequential presentation, but mandatory procedure requires that the witness must watch the entire lineup twice before making any identification. Valentine, Darling and Memon (2007) asked whether the effectiveness of video lineups may be improved by changing the instruction to require a yes/no response to each face as it is presented (Lindsay, Lea, & Fulford, 1991). Participants had witnessed a staged live incident about a week prior to viewing a video lineup constructed by the police using a national police database to select lineup membe
rs. The existing (view twice) procedure was compared to the strict sequential instructions described by Lindsay et al., 1991). When the perpetrator was present in the lineup, he was more likely to identified using the current “view twice” procedure (65% compared to 36%). When an innocent suspect was in the lineup, 23% of witnesses mistakenly identified a foil following the “view twice” instruction compared to 10% under the strict sequential instruction, but this difference was not statistically reliable. In conclusion, the strict sequential instruction significantly reduced the sensitivity of video lineup but the reduction in foil identifications was not statistically significant.
15.7 MALLEABILITY OF WITNESS CONFIDENCE
An important research finding is that witness confidence is changeable and is influenced by information that the witness acquires after attending an identification procedure. Receiving feedback that the witness identified the suspect or that somebody else made the same identification will increase the witness’s confidence in their identification (see Case Study 15.2). Not only does confirming feedback tend to make the witness subsequently more confident in their identification, but it also tends to inflate estimates of a range of subsequent testimony, including: how long the culprit was seen for; how close they were; and how much attention the witness paid (Wells & Bradfield, 1998). Furthermore, confirming post-identification feedback tends to make eyewitnesses over-confident; that is, they now express more confidence in their identification than is warranted (Semmler, Brewer, & Wells, 2004). Wright and Skagerberg (2007) showed that feedback affected the confidence of witnesses and victims of real crimes. By the time a witness gives evidence in court they are likely to have received or inferred confirming feedback, if only by virtue of the fact that they have been called upon to give evidence. The witness is less likely to be asked to attend court if they have identified the “wrong” person.
CASE STUDY 15.2 THE MURDER OF JILL DANDO
Jill Dando was a famous British TV presenter. On 26 April 1999 she returned to her home at about 11.30am. She was killed on her doorstep by a single shot from a handgun. Nobody saw the murder, but two neighbours who heard a muffled cry looked out of their window and saw a man walking away from the house but neither realised at the time that there was anything wrong. It was accepted that the man seen by these two witnesses was undoubtedly the murderer. Initially, the police suggested that the murder had been carried out by an assassin but a year later, Barry George, a local unemployed man, was charged with the murder.
Identification evidence was central to the case. Sixteen people who saw a man in Gowan Avenue either on the morning of the murder or the previous day, attended an identification procedure. The first five witnesses saw a live parade and the remainder a video parade after George was advised by his solicitor not to attend any further live parades. Barry George had been clean-shaven at the time of the murder but now had a beard. Therefore, all of the men in the lineup had beards. Neither of Jill Dando’s neighbours identified George. He was positively identified by only one witness. This witness had seen the face of a man for five to six seconds some hours before the murder and made her identification 17 months later.
After making her identification, the identifying witness was given a lift home in a police car with two other witnesses. During the journey the other witnesses discovered the positive identification she had made, and apparently influenced by this knowledge, both subsequently made statements that they too would have identified George, but were hindered by not being able to see his build and height in the video.
The case set a legal precedent because identification evidence was admitted from four witnesses who did not make a positive identification at the lineup. In addition to the two retrospective identifications, a postman stated that the man he saw was not in the lineup, but he recognised a man he had had a conversation with about the Dando case in the days after the murder, a fact confirmed by Barry George. The prosecution argued that there was an underlying unity of the description of the man these witnesses saw, which pointed to George as the murderer.
Barry George was convicted of murder in July 2001 by a majority verdict and sentenced to life imprisonment. In July 2002 the Court of Appeal upheld the principle that a witness who did not identify the suspect can still give evidence in court (R v George, 2002). However, in May 2007 the Criminal Cases Review Commission referred Barry George’s conviction back to the Court of Appeal on the single ground that new evidence called into question the significance attached to the firearms discharge evidence at trial (R v George, 2007). His conviction was quashed in November 2007.
15.8 OFFICIAL GUIDANCE
The conduct of identification procedures in England and Wales is governed by Code D of the Police and Criminal Evidence Act (PACE) (1984) Codes of Practice. The current code came into force in 2017. The major provisions of the code are as follows. If an identification is disputed, a video identification procedure containing moving images must be offered unless it is not practicable or a “live” identity parade is considered more suitable. In addition, all lineups should consist of a minimum of eight foils and one suspect; the foils must “resemble the suspect in age, general appearance and position in life”. Witnesses must be advised that the person they saw may not be present and must view the entire lineup at least twice. The person who runs the procedure should not be involved in the investigation of the case. The suspect has the right to a legal representative to be present. The suspect and the legal representative may object to the procedure (e.g. the selection of foils) and their reason for objection must be recorded. If the witness has previously been shown photographs, details of the photographs shown should be recorded. Anything the witness says should be written down before leaving the identification room.
In the United States identification procedures are regulated at state level or below, so there is no one federal code of practice2. Identification from photographs is commonplace. Procedures are often conducted by an investigating police officer. The US Department of Justice has issued a guide on identification procedures, but it does not have the force of law (Technical Working Group for Eyewitness Evidence, 1999). The guide makes recommendations for sequential and simultaneous lineups of photographs or “live” volunteers, but does not endorse one method over another. A minimum of five foils is required. Advice is provided on selecting foils so that the suspect does not unfairly stand out, giving unbiased instructions and carefully recording all identifications or non-identifications. A landmark judgement in the New Jersey Supreme Court was strongly influenced by the US Department of Justice guide (New Jersey v. Henderson, 2011). Model instructions to the jury were required to be drawn up by the Henderson judgement.3 A recent US National Research Council report reviewed identification procedures from both a scientific and legal perspective. The report surveyed the scientific literature to identify gaps and assessed what could be learned from other fields, and made recommendations for best practice (National Research Council, 2014; see Valentine & Fitzgerald, 2016 for a review).
15.9 IDENTIFICATION FROM CCTV
In view of the fallibility of eyewitnesses, it may be a comforting thought that in many cases we can rely on CCTV images to identify perpetrators, thus cutting out the human error involved in eyewitness testimony. In the UK in particular, this idea has proved highly attractive to politicians and the public alike. The UK is believed to have the highest density of CCTV surveillance in the world. (See Norris, McCahill, & Wood, 2004 for a review.) CCTV does have many benefits including its use in the investigation of crime. However, identification of perpetrators from CCTV has proved to be surprisingly susceptible to human error.
People are extremely good at recognising highly familiar individuals (e.g. work colleagues, friends and family) even from low quality images. Familiar people can be recognised from poor quality CCTV images with over 90% accuracy (Bruce, Henderson, Newman, & Burton, 2001; Burton, Wilson, Cowan, & Bruce, 1999). However, we are surprisingly poor at matching images, taken by different cameras, of an o
therwise unfamiliar person. For example, Bruce, Henderson, Greenwood, Hancock, Burton and Miller (1999) asked participants to choose the face from an array of ten high-quality photographs that they thought matched a target face. The photograph of the target was taken from a studio video recording made on the same day as the still photographs in the array were taken. The set of faces consisted of 120 young, male, clean-shaven Caucasian police trainees, supplied by the Home Office. The arrays consisted of faces judged to resemble each other so that selections could not be based on substantial differences in hairstyle, weight or age. When the viewpoint and facial expression of the target and the images in the array matched (i.e. a comparison was made under ideal conditions), participants only made the correct selection in 79% of the arrays. Bruce et al. concluded, “The implication of these findings is that courts must be aware that caution should be used when the impressions of resemblance are used to establish the identity of unfamiliar people, even when the quality of video tape is high” (emphasis as in the original). Henderson, Bruce, and Burton (2001) found similarly high error rates in identifying unfamiliar persons from high-quality video images with a two-face array.
The forensic implications of a separate study of recognition of faces from a poor-quality video, typical of that recorded by a commercial security CCTV system, was summarised as follows: “identification of these types of video sequences is very unreliable, unless the viewer happens to know the target person” (Burton et al., 1999). Burton et al.’s study showed that police officers, with experience in forensic identification and an average of over 13 years’ service, perform as poorly as other participants unfamiliar with the targets. A police officer is no more likely to correctly identify somebody from video than anybody else who has a similar level of familiarity with the target.