I

Eyewitness Identification Strategies in Simultaneous and Sequential Lineups

A Dissertation Proposal

8/14/02

Heather D. Flowe

Abstract

The American Psychology-Law Society recently issued guidelines for conducting criminal identifications. Investigators are encouraged to construct lineups using distracter faces that match witness descriptions of the perpetrator. Additionally, presenting the photographs sequentially for identification is preferred over using photospreads. These recommendations are based on the assumption that witnesses use a relative decision strategy to evaluate simultaneously presented faces. The design of the present project allows for several critical tests of the relative judgement hypothesis. A signal detection analysis of simultaneous and sequential lineups is presented, which suggests the rate at which false identifications are reduced depends on criterion placement and lineup member similarity. The proposed project will systematically vary the featural similarity of faces and present them simultaneously or sequentially to witnesses for identification. Criterion-based manipulations include admonishment strength, witness perception of task difficulty, and instructions that encourage witnesses to analyze faces using either a holistic or a featural judgement strategy. Accuracy, eye movements, response times, and decision confidence data will be collected to describe the judgement strategies used in lineup identification.

I. Introduction

The American Psychological Society recently celebrated the application of psychological research findings to the legal system, proclaiming that empirical findings and psychological theory provide a "far better foundation for personal decisions and public policies than does intuition or standard practice" (APS 2001 Convention Program, p. 35). Indeed, much psychological research has been devised over the last 25 years to demonstrate the conditions under which eyewitnesses might be prone to making identification errors. Psychologists have used the results of this laboratory research to persuade the legal system to make procedural changes aimed at reducing the rate of mistaken identification. In particular, psychologists have criticized the use of the photographic simultaneous lineups (e.g., Lindsay & Wells, 1985; Wells et al. 1998), which might be the most frequently used identification procedure in trial cases involving eyewitnesses (Flowe & Ebbesen, 2002).

In a simultaneous lineup, witnesses are shown an array of photographs, which include the suspect along with persons known by the police to be innocent (foils). Lindsay and Wells (1985) have argued that the simultaneous procedure does not adequately protect innocent suspects from being mistakenly identified. They hypothesize that simultaneously displaying the photographs encourages witnesses to choose the person in the lineup who looks relatively the most similar to the suspect. If most innocent suspects appearing in police lineups bear a greater resemblance to the culprit than do the foils, then a relative judgement process will lead to many mistaken identifications. Lindsay and Wells (1985) proposed that the sequential lineup procedure is a better alternative to the simultaneous lineup. In a sequential lineup, photographs are presented to witnesses one at a time. The next picture is displayed if the photo is rejected, and once a photo has been rejected, the witness is not allowed to see it again. The procedure continues until the witness identifies a photograph as the culprit. No other photos are presented once a photo has been identified. Lindsay and Wells (1985) maintain that displaying the pictures sequentially prevents witnesses from making relative comparisons among lineup members. Compared to witnesses viewing a simultaneous lineup, witnesses who see the pictures sequentially will base their decision on their memory for the culprit, rather than on relative similarity.

Research has found that by switching from a simultaneous to a sequential procedure, false alarms are reduced while hit rates are not appreciably affected (Cutler & Penrod, 1988; Levi, 1998; Lindsay, Lea, & Fulford, 1991; Lindsay & Wells, 1985; Lindsay, 1999; Lindsay, Pozzulo, Craig, & Lee, 1997; Parker & Ryan, 1993; Sporer, 1993). Since it seems as though the procedure will protect innocent suspects while still allowing for the identification of guilty suspects, psychologists have recommended that police investigators conduct lineups sequentially. In an official position paper, the American Psychology Law Society recently has argued that sequential lineups protect innocent suspects while allowing for the conviction of the guilty (Wells et al. 1998). The paper also aggressively argues that sufficient research evidence exists to support the idea that witnesses are prone to making relative judgements in simultaneous lineups. The arguments put forth were instrumental in leading New Jersey to adopt the sequential lineup procedure exclusively. In advocating for the use of sequential lineups, Attorney General John Farmer reported that scientific studies have "proven that witnesses have a tendency to compare one member of a lineup to another, making relative judgments about which individual looks most like the perpetrator" (http://www.state.nj.us/lps/dcj/agguide/photoid.pdf).

The purpose of this thesis is to more closely examine the decision strategies that witnesses use when confronted with simultaneous and sequential lineups. No experiment to date has demonstrated that witnesses mainly make relative comparisons when evaluating a simultaneous lineup or absolute judgements when evaluating a sequential lineup. Furthermore, the research evidence that has been cited as consistent with the relative judgement hypothesis is rather indirect: Since the false alarm rate tends to be higher for simultaneous lineups compared to sequential lineups, theorists such as Wells have inferred that eyewitnesses must be making relative judgments. The proposed research will describe the strategies that witnesses use to inspect pictures in simultaneous and sequential lineups by measuring eye movements, decision confidence, and response times in addition to accuracy rates.

Another problem under consideration is that lower false alarm rates are not always observed in sequential compared to simultaneous lineups (e.g., Hiatt & Shapiro, 2001; Perlberg, 2001). One explanation for these inconsistent results is that the similarity of the lineup members relative to the culprit varies across studies. Researchers seldom measure lineup member similarity in their experiments. In a recent review of the eyewitness identification literature, only 1% of the studies formally measured the similarity of the foils to the culprit (Flowe & Ebbesen, 2001a). The remaining studies selected foils and the innocent suspect based on the experimenter's judgement that they were similar enough in appearance to the culprit. Therefore, the present project also aims to systematically examine how the featural similarity of lineup members influences the choices made by witnesses. Before describing the research methodology that will be employed, an overview of relevant research is provided, including the evidence cited in support of the relative judgement hypothesis. An alternative model of identification behavior will be presented and the procedures that have been proposed for constructing fair lineups will be reviewed.

The Relative Judgement Hypothesis

Wells et al. (1998) cited the "removal without replacement" procedure as the best evidence that eyewitnesses use a relative judgment strategy in simultaneous lineups. The removal without replacement procedure involves two parts. After viewing a staged crime, eyewitnesses are shown a culprit-present simultaneous lineup. The distribution of choices over all available alternatives in the lineup and the rate of rejecting the lineup are both recorded. In the second part, another group of eyewitnesses undergoes similar experimental procedures, except that the target's picture has been removed from the lineup and not replaced with any other photograph. According to Wells et al. (1998), if a relative judgment strategy is being used, then eyewitnesses viewing the target-removed lineup will be inclined to select the most familiar face from the remaining set of foils instead of correctly rejecting the lineup. If witnesses were using an absolute judgment strategy, then they would correctly reject the lineup because the culprit is not there. As shown in Figure 1, Wells (1993) found that the subjects were more likely to pick a foil and less likely to reject the lineup after the target had been removed. Since the most frequently identified foil in part one of the experiment was most often identified in part two, Wells inferred that witnesses simply identified the relatively most familiar face.

A second source of evidence that relative judgments are made is the effect that biased lineup instructions have on identification decisions (Wells et al., 1998). Eyewitnesses who are admonished that the culprit "may or may not be present" prior to viewing a target absent simultaneous lineup are less apt to make false identifications than eyewitnesses who are not given such a suggestion. In a recent meta-analysis, Steblay (1997) found admonishing witnesses that the culprit might not be in the lineup increases the rate at which target absent lineups are rejected while having "minimal" effect on the rate at which the culprit is correctly identified. That is, admonishment to use a stricter criterion only seems to affect errors made to target absent lineups. One interpretation of this outcome is that admonishment discourages eyewitnesses from making relative judgments (Wells, et al. 1998). Presumably, if witnesses are led to believe that the culprit is in the lineup, they select the most familiar alternative. Witnesses who do not possess such a belief avoid making relative judgments, basing their decision on match-to-memory evidence instead.

Another research finding used to support the contention that relative judgment is used in simultaneous lineups is the effect that the dual lineup procedure has on accuracy rates. In the dual lineup procedure, eyewitnesses are first shown a blank lineup (i.e., a lineup that does not contain the suspect) before the actual lineup test. Wells (1984) found that participants who rejected the blank lineup, compared to those who picked someone out, were less likely to false alarm on a subsequently presented target absent lineup. Hit rates, however, did not differ depending on whether the participant chose someone from the blank lineup. Wells and his colleagues (1998) argued that blank lineups might be used to screen out witnesses who are prone to making relative judgments.

In addition to the experimental research, self-report data has been used to investigate lineup judgment strategies (Lindsay, Lea, Nosworthy, 1991; Lindsay & Bellinger, 1999). Subjects tend to agree that they use a relative strategy when viewing a simultaneous lineup and an absolute strategy when viewing a sequential lineup. Furthermore, these studies find that people are more accurate if they report using an absolute rather than a relative strategy. Some subjects, however, have been known to report using an absolute strategy even though the experimenter observed them comparing lineup pictures (Lindsay, & Bellinger, 1999). Equally important, the factors that control self-reports of mental processes might not affect the actual decision processes. For example, subjects might be report more often that they compare pictures in a simultaneous lineup because they can shift their visual gaze from one picture to another and cannot do so in a sequential lineup. Gaze shifting might have little to do with whether subjects are using an absolute similarity standard. Additionally, relative comparisons might also be made in sequential lineups, with witnesses mentally comparing each photo to previously seen ones.

Rather than simply asking witnesses whether they used an absolute or a relative strategy in making their decision, more direct measures, such as recordings of eye movements, might be more informative in describing differences in how eyewitnesses inspect photographs presented to them in simultaneous and sequential lineups. In viewing a simultaneous lineup, for example, it is conceivable that witnesses might employ both relative and absolute strategies, but at different times over the course of the identification task. They first might examine all of the members of the lineup, eliminating those that do not surpass the criterion they have set for selection. In other words, those that do not appear similar enough to the culprit are immediately ruled out. Relative comparisons among lineup members might then be made to determine which member is globally the most similar to the culprit in memory. For instance, there is some evidence to suggest that holistic processing (i.e., analyzing combinations of features, rather than single features one at a time) is relied upon more often in processing familiar rather than unfamiliar stimuli (Tanaka & Farah, 1993; Rhodes, Brake, Tan, & Taylor, 1989). Holistic processing might also increase the discriminability of faces that are similar (Haig, 1984).

Additionally, both presentation procedures might involve absolute judgements to some degree, in that specific facial features, such as the eyes/eyebrows and hair, which are important in face recognition (Haig, 1984), are evaluated for each member. Given that in sequential lineups the pictures are presented one at a time and the witness knows that the picture is going to be displayed only once, specific facial features might be analyzed individually more often in sequential compared to simultaneous lineups. Though facial features could be analyzed individually more often in judging faces presented sequentially, overall familiarity, or a more holistic process, probably also plays a role (Cottrell, Dailey, Padgett, & Adolphs, 2001). Furthermore, the order in which witnesses employ relative and absolute strategies might differ in simultaneous and sequential lineups, thereby influencing identifications. The research proposed in this project will be the first to examine how the judgement process changes as a function of lineup member similarity and the number of photographs that are displayed.

In addition to the possibility that witnesses use a combination of strategies, another issue to consider in evaluating the relative judgment hypothesis is that witnesses often do not select anyone from simultaneous lineups. If witnesses were simply selecting the relatively most familiar person, then they should almost always select someone. Eyewitness identification researchers and the police do not randomly select pictures to compose a lineup. Instead, photos are chosen on the basis of their match to the suspect's photo or the culprit's description (Flowe and Ebbesen, 2001b). As such, there should always be at least one member that looks similar to the culprit. Witnesses who make relative judgements should identify someone from the lineup as the suspect more often than they would reject the entire lineup. However, a recent meta-analysis of simultaneous lineup experiments found that 50% of the time, laboratory witnesses reject the lineup (Ebbesen & Flowe, 2001). Additionally, in real world police investigations, witnesses fail to select anyone from a simultaneous lineup 50% of the time (Behrman & Davey, 2001). These rejection rates suggest that witnesses are using an absolute judgement process of sorts. If they were judging the lineup only on the basis of relative similarity, then it seems as though they would select someone more often than 50% of the time.

There is also an alternative explanation for the finding that witnesses false alarm more often in simultaneous compared to sequential lineups. An additional feature that differentiates sequential lineup presentation is that witnesses do not know the number of photographs they will be seeing. Instead, a stack of photographs is displayed to give the impression that many will be shown. This procedural detail was added in order to prevent people from picking the last photograph simply because they know no others will be displayed. The identification session ends after a predetermined number of pictures (usually 6) have been presented. One potential effect of this procedure, however, is that witnesses will withhold making a choice, concerned that a better match to the suspect is still to come. In other words, witnesses viewing sequential lineups might set their decision criteria higher than those viewing simultaneous lineups might.

Applying signal detection theory to this problem allows for several predictions regarding the effect on hit and false alarm rates that results from a switching to a sequential procedure over a simultaneous procedure (see Ebbesen & Flowe, 2001 for a more in-depth look). Figure 2 shows foil and culprit distributions for a target present lineup. For simplicity, the example assumes that a 3 member lineup, containing two foils and the culprit, is being shown to eyewitnesses. The underlying model of signal detection theory consists of two normal distributions, one representing the signal and the other noise. In the current application of signal detection theory, the signal distribution is the culprit distribution, and the foil (and the innocent suspect look-a-like if the culprit is not in the lineup) distributions are the noise distributions. In most signal detection applications, the distributions are aligned on a strength of memory to evidence dimension. For the present purpose, the dimension of interest is familiarity, which is being conceived of as the lineup member's similarity to the culprit in memory. The application of signal detection theory to event memory research on lineup identification also differs from traditional applications in other respects. For instance, in face recognition research, subjects learn a list of faces and then "old" and "new" faces are presented at test. The distributions represent the relative familiarity of old (target) and new (distracter) faces within the head of a single subject. In event memory studies, subjects see the same lineup. As a consequence, the target and foil distributions represent the familiarity of the same few photos over many subjects. Second, since it is conceivable that subjects vary in where they place their criterion, in applying signal detection theory to event memory studies, the criterion should be treated as an average criterion over subjects.

The top panel of Figure 2 illustrates that if switching to sequential lineups causes witnesses to place their criterion at a higher level, then the rate at which guilty suspects will be identified will be reduced. Though Wells et al. (1998) have argued that hit rates do not differ between simultaneous and sequential lineups, there is evidence that hit rates might be affected. Ebbesen and Flowe (2001) found in a meta-analysis of simultaneous and sequential lineup experiments that witnesses were slightly better at finding the culprit in a simultaneous lineup. Figure 2 also shows the effect of moving to sequential lineups on false alarms. If a sequential lineup causes witnesses to shift their criterion to a higher level, then witnesses will be less likely to choose an innocent suspect if the pictures are presented one at a time (see bottom panel of Figure 2). Furthermore, the effect on false alarms will be greater than the effect on hit rates as a result of switching procedures. This prediction is consistent with the research conducted to date, which shows that the sequential lineup affects false alarms more than hits (Ebbesen & Flowe, 2001).

An important issue from an applied point of view that emerges from this analysis is the rate at which guilty suspects appear in police lineups. If most often guilty persons are placed in lineups, then switching to a sequential lineup will result in more guilty suspects going free. On the other hand, if most people appearing in police lineups are innocent, then switching to a procedure that raises the witness' decision criterion would seem like a good measure to take. Deciding which procedure would best serve the interests of justice is further complicated by the question of where real world witnesses place their decision criteria. If placement is high, guilty persons that would have had a higher probability of being chosen had a simultaneous lineup been used will go free.

Extending the signal detection theory application to instances in which the lineup members are more or less similar to the culprit leads to some further predictions that are relevant if one wishes to generalize laboratory research to real world lineups. Displayed in Figure 3 is a signal detection analysis of 3 lineups that vary in the similarity of the foils relative to the culprit. The similarity of the culprit in memory relative to the culprit's photo is held constant across the 3 lineup conditions. As shown, increasing the similarity of the foils relative to the culprit results in a greater number of foil choices in both simultaneous and sequential lineups. Foil choices increase at a greater rate in simultaneous lineups. Culprit choices, however, should not increase over the similarity conditions because his familiarity is being held constant. In target absent lineups, the likelihood of selecting a foil increases as similarity rises (see Figure 4). The effect should be more pronounced for simultaneous lineups. Choices of the innocent look-a-like should not increase with foil similarity, since his likeness to the culprit is not changing. If his similarity to the culprit were increased along with the foils, then the likelihood that he would be selected would increase as well.

On the other hand, if eyewitnesses use an absolute strategy in evaluating sequential lineups, then increasing the similarity of the foils (or the innocent suspect) should not increase choice rates in such a fashion. Wells et al. (1998) argued that the relative judgement strategy dictates that the rate of rejecting the lineup will not be affected by whether one or all of the foils resemble the suspect. Instead, foil choices will be redistributed when the similarity of lineup members relative to the suspect is altered, with witnesses selecting the member that appears most similar to the culprit. The proposed experiments will test this competing hypothesis by manipulating the similarity of the foils relative to the target and innocent suspect to examine whether the hit and false alarm rates vary according to the similarity of lineup members. Strength of admonishment will also be experimentally manipulated to determine whether the rate at which both guilty and innocent suspects are identified in simultaneous and sequential lineups is affected.

The amount by which false alarms are reduced in the real world as a result of switching to sequential lineups might depend on how police construct lineups. If police typically construct lineups in which the members possess a low degree of similarity, then the signal detection analysis predicts that the rate at which false alarms are reduced by sequential lineups will be lower than if lineups typically contain members that are highly similar, and vice versa. Therefore, the proposed project will begin by measuring suspect and foil similarity in a random sample of police lineups (Experiment 1). An additional purpose for doing so is to compare the similarity of the police lineups to those that will be constructed by manipulating specific facial features (Experiments 2-4).

Methods Used to Construct Lineups

If an identification is to have any probative value at all, obviously the witness should not know which lineup member the police suspects. For instance, an extreme version of a lineup that is unduly suggestive is the "Oklahoma Showup", in which the police "accidentally" parade the shackled suspect in front of the witness prior to conducting the lineup (Wall, 1956). A less overt but nonetheless suggestive method is to place the suspect in a group of foils that are in appearance dissimilar from the suspect. Gross forms of this practice have long been recognized by courts as being unfair, such as in cases where a suspect is placed in lineup among people of a different race (e.g., People v. Seppi, 1917). In other instances, practice has been to conceal distinguishing features, such as covering the foot of everyone in the lineup with a rug if the suspect has a clubbed foot, making everyone wear glasses if the suspect wears them, or covering everyone's eye if the suspect wears a patch (Williams, 1955). Procedures such as these prevent the suspect from immediately "popping out" of the lineup. On the other hand, the lineup members cannot be so similar that witnesses who have a memorable and accurate image of the perpetrator will be unable to make a correct identification if he is present in the lineup.

A national survey of police officers found that lineups are typically constructed by finding foils that match the suspect's photograph (Wogalter, Burger, & Malpass, 1993). Luus and Wells (1991) argue that the match-suspect strategy does not allow one to determine the type, number, or degree of featural overlap necessary for constructing a fair lineup. The purpose of having foils in the first place is to control for the occurrence of identifications made by chance. In order to serve their function, Luus and Wells (1991) argue that witnesses should not be able to use deductive reasoning to eliminate foil choices that clearly violate their memory for the perpetrator, such as in a situation where the witness knows that the culprit has brown hair and is 5' 3", yet one of the foils is 6' 0" and has blonde hair. They further argue that the features that could be used to select foils equal in similarity to the suspect's photo are numerous. Carried to the extreme, lineup foils selected by matching as many features as possible will result in a lineup with members that are virtually identical, a problem they called the "clone anomaly". Luus and Wells (1991) contend that selecting foils on the basis of matching only the features mentioned in the witness' description is a more practical method because it delimits the list of potential features that could be used to create a lineup. They suggest that lineup foils should be selected from a pool of persons that fit the "free recall" description of the culprit provided by the eyewitness. Furthermore, the people chosen from this delimited pool to comprise the lineup should be selected such that they are maximally different from the suspect, such that "propitious heterogeneity" in the lineup is achieved.

Navon (1992) also advocates for constructing lineups based on matching the foils to the culprit's description. He used probability theory to argue that the match-suspect strategy could lead to the creation of biased lineups in which the innocent suspect looks more like the actual culprit than do any of the foils. Police select foils for the lineup based on a different criterion than the suspect. The innocent suspect is found by the police on the basis of his match to the witness' description of the culprit, whereas the pictures selected to serve as foils are chosen on the basis of their match to the suspect's photo. As a consequence, Navon (1992) argued that witnesses would be more likely to select the innocent suspect over any of the foils because the suspect shares more features in common with the perpetrator. Clark (1999) found support for Navon's predictions using a computer-simulated model of eyewitness identification.

Wells, Rydell, and Seelau (1993) experimentally tested these predictions by presenting witnesses with a videotaped crime scenario followed by an identification task. The lineups were constructed by matching the foils to the perpetrator's photo, matching the foils to the perpetrator's description, or by mismatching the photos to the perpetrator's description. The pool of foils used to construct the match-description lineups fit the general description of the culprit, but otherwise did not resemble him or her. The results showed that the match-description method produced a higher rate of hits compared to the match-suspect method. Contrary to the prediction made by Navon (1992), false alarm rates between the two construction procedures did not differ (and were not in the expected direction, with the false alarm rates being .20 and .28, for suspect-matched and description-matched, respectively). The mismatch condition had the highest rate of both hits and false alarms. Based on these data, Wells et al. (1998) recommended that the match-description method be used to create fair lineups. The argument they advanced is that the match-description method allows for sufficient variation among lineup members such that witnesses with accurate memories can distinguish the culprit from the others when he or she is present. Innocent suspects will also be protected because they will be no more likely chosen by the witness than any other member will.

Tunnicliff and Clark (2000), however, failed to replicate the finding that the match-description procedure produces a greater number of hits than matching the suspect's photo. Foil to suspect similarity judgements made by a group of independent raters confirmed that similarity was higher on average for the match-suspect group compared to the match-description group. Even so, in the two attempts that were made to replicate the description-matched advantage (police officers constructed the lineups in the first attempt and college students constructed the lineups in the second), correct and false identification rates did not differ depending on construction technique. Additionally, though not reliably so, subjects were somewhat more likely to incorrectly identify the innocent suspect in the description-matched rather than in the suspect-matched condition. No-pick responses did not significantly differ in the match-description condition depending on whether the culprit was present or not. For suspect-matched lineups, however, subjects were more likely to pick somebody if the culprit was present rather than absent.

Moreover, if lineups are constructed using the match-description method, the police still might have to compare the similarity of photos to each other in order to determine whether they have created a fair lineup. To be sure, the police should not just randomly select photos from a database search of the features described by the witness and put them into a lineup. To illustrate some of the complications that could arise, suppose a witness in Iowa describes the culprit as having black hair and blue eyes, and these are the only features recounted. A search of the Iowa Sex Offender Registry using black hair and various eye colors yielded (out of a total number of 1441 database records): 244 registrants with black hair and brown eyes, 13 with black hair and blue eyes, 1 with black hair and green eyes and 1 with black hair and hazel eyes (and he was almost bald and wearing an eye patch). To create a fair lineup, the police would most likely need to draw foils from a larger pool of photos and check to see that the similarity of the lineup members is sufficient to some degree. For instance, Wogalter, Marwitz and Leonard (1992) found that lineup construction methods that incorporate foil to foil similarity produce fairer lineups than those constructed on the basis of matching the description alone.

The police in the above example might also have to prompt the witness for additional descriptors. Lindsay, Martin and Webber (1994) questioned whether witnesses providing only a few features or vague descriptions would lead to the creation of lineups that are biased against the innocent suspect. They analyzed descriptions given by laboratory witnesses and found they were often lacking in detail (e.g., hair color is described as being "light" or "dark"), and that witnesses often left out information that they must have noticed, such as the perpetrator's sex and race. Similar results were obtained in an analysis of eyewitness descriptions of culprits appearing in newspapers. Lindsay et al. (1994) also took issue with Luus and Wells' (1991) recommendation that the foils, selected based on their match to the description of the culprit, should not otherwise resemble the suspect. They argued that the police could select foils that were highly dissimilar to the innocent suspect, leaving him or her to stand out in the lineup. To test this idea, they selected distracters that were either highly similar to the suspect (match-suspect strategy) or who varied substantially in their similarity to him and to each other (match-description strategy). A third type of photo array was created by choosing foils that were rated the least similar to the culprit but who matched his description (biased match-description). In the target absent lineups, the innocent suspect was moderate in his degree of similarity to the perpetrator. When the suspect was present, hit rates did not differ depending on the lineup construction technique used, though the means were in the expected direction, with .66 for match-suspect, .79 for match-description, and .81 for biased match-description. False alarms also did not differ in the match-suspect compared to the match-description condition (.08 versus .25). However, in the biased match-description condition, a higher rate of selecting the innocent suspect was obtained (M = .50) than in the match-suspect or match description conditions. Even when the photographs were presented sequentially, false alarms were higher in the biased match-description lineups (M = .25) than in the match-description (M = .03) or match suspect sequential (M = .00) lineups. Although they concluded that the match-description method should not be abandoned, they called for additional research to determine what degree of descriptive detail was needed in order to produce fair lineups.

Despite the fact that only two studies had been conducted at the time (and now recently published findings are conflicting), the Executive Committee of the American Psychology/Law Society ruled that "selecting the distracters so as to resemble the suspect is not a desirable practice" (Rule 3, Wells et al. 1998). They further ruled that lineup distracters be chosen on the basis of their match to the perpetrator's description. This set of recommendations has been incorporated into practice guidelines instituted by the National Institute of Justice (1999) for the collection of eyewitness evidence: "Select fillers who generally fit the witness' description of the perpetrator. When there is a limited/inadequate description of the perpetrator provided by the witness, or when the description of the perpetrator differs significantly from the appearance of the suspect, fillers should resemble the suspect in significant features." The recommendations, however, are silent as to how many features should be considered inadequate, which features are significant, or to what degree the foils should resemble the suspect in order to produce a fair lineup.

The proposed project will begin to explore these issues. Since previous research on facial recognition accuracy has found that head outline, the eye/eyebrow combination (Haig, 1984), and hair (Cutler, Penrod & Martens (1987) are important features for discriminating "old" from "new" faces, the similarity of the suspect to the foils using these features will be systematically controlled using composite drawings of faces (Experiment 2). Because eye movements are being recorded, other facial features that seem to play a role in discriminating the culprit (or innocent suspect) from the foils will also be noted to guide future investigations. Presentation procedure (simultaneous or sequential) and admonishment strength will also be manipulated. Other manipulations include the effects of lineup similarity and the number of pictures that witnesses expect to see in order to determine whether criterion placement is affected (Experiment 3). Finally, an attempt will be made to extend the findings to real photos of both college students and criminal offenders. Lineups will be created by matching either the description of the target or the photograph of the target using a predetermined number and type of features (Experiment 4).

In all of the experiments, administration of the lineups will be computerized. Additionally, the likeness of the culprit to his lineup picture will be held constant. Previous studies have not controlled for this factor. Differing hit rates across studies are difficult to interpret because having a target photo that is more or less similar in appearance to the culprit shown during the study period might affect correct identifications.

In the interest of the reader's time, the Appendix summarizes the predictions that this project will test.

II. Measuring Fairness in a Random Sample of Police Lineups

A random sample of police lineups (N = 10) was collected from the case archives of the San Diego District Attorney's Office. A description for each suspect was derived from eyewitness statements to the police. Mock witnesses (N = 48) viewed the lineups and were told to select the one person who they thought committed the crime. The mock witness procedure is used by researchers to determine whether a lineup is fair (see Malpass & Lindsay, 1999 for a review). If the suspect is selected at greater than chance levels, then the lineup is considered to be biased against the suspect. As a method for simulating a witness' memory for the culprit's descriptive features, the culprit's description was given to 23 of the witnesses. Following their identification, witnesses answered an open-ended question asking them to explain what motivated their choice.

The similarity of the lineup members was rated by an independent group of subjects using a 100-point scale, anchored at "Not at all Similar" and "Completely Alike". Ratings were made pair-wise within each lineup. The lineup pairs were displayed in random order across the 10 lineups on a computer screen. The scores were averaged to form measures of the similarity of each foil to his respective culprit, the similarity of the foils to each other, and the similarity of the suspect to the foils. The photos were also combined across the lineups and presented one at a time in random order to an additional group of subjects who rated either how "criminal" or how "distinctive" each of the faces was.

Mock witnesses at a rate higher than expected by chance selected 2 of the suspects only when given a description. Three of the suspects were selected more often than chance expectation only when a description was not provided. Of the remaining suspects, 2 were selected whether or not a description was given, and 3 were never selected at a rate above chance. Conditioning on whether or not a description was provided, none of the similarity measures predicted suspect choice rates, choice rates within the sample of photos overall, or choice rankings within the lineups. However, compared to the other suspects, the suspects that were selected regardless of whether or not a description was given (n = 2) tended to be rated as less similar to the foils than the foils were to each other, though the effect was not significant (p = .11). Additionally, if witnesses were not given a description of the culprit, the member most frequently selected was the one with the highest criminality rating in 7 of the lineups. Distinctiveness did not predict mock witness choices.

A preliminary analysis of the self-report data indicated that witnesses who picked the suspect out of the lineup most often used hair, build and age. In some instances, descriptors related to the suspect's demeanor (e.g., the suspect was "dirty", "violent", or "dressed well") reportedly guided witness choices. Closer inspection of the lineups suggested that 6 of the suspects may have stood apart from the others in appearance because the context of the lineup made the suspect more noticeable than the foils (i.e., the suspect was looking in a different direction, had a different emotional expression, or he had his head tilted). None of the witnesses reported picking the suspects due to these reasons.

Overall, the foils were rated as moderately similar to the suspect across lineups (M = 55.88, SD = 6.63, Range = 46.15 - 66.99). The results of this study suggest that similarity needs to be measured at the level of individual facial features in order to predict the ease with which a suspect will be chosen from a police lineup. Additionally, two pilot studies in which lineups were constructed based on similarity ratings indicated that choice rates did not vary depending on whether the lineup foils were low or high in average similarity to the culprit. Therefore, the next experiment will systematically manipulate facial features to form lineups of varying degrees of similarity.

III. Experimentally Manipulating Facial Features

To investigate choice rates as a function of featural similarity and lineup presentation procedure, subjects will be presented with a composite drawing of a male face for 10 seconds and attempt to identify him from a 6 person lineup following a 4 min distracter period. The experiment will be conducted in two phases, with strength of admonishment (weak or strong) varied across phases. Lineup presentation procedure (simultaneous or sequential) will be controlled between subjects. Participants will expect to view 6 pictures in both the simultaneous and sequential lineup conditions. For each culprit, participants will be tested with both a target present and a target absent lineup, with order of presentation randomly determined.

FACES 4.0, a composite drawing program developed for law enforcement, will be used to construct the culprit, the innocent suspect, and the foils. The target picture in the lineup will always be identical to the picture that was studied. For target absent lineups, the innocent suspect will always share 3 features (hair, face shape and eye/eyebrows) in common with the suspect. The position of the culprit or innocent suspect in each lineup will be randomly determined for each participant.

Lineup foils will be randomly drawn from a total of 400 unique faces. The foils will be manipulated such that the hair, face shape, eyes/eyebrows, or no feature is identical to that of the culprit. Combinations of the features will be used to control the number of features foils share in common with the culprit (8 combinations, including the no feature match condition, are possible). All of the foils in a given lineup condition will have the same feature(s) in common with the suspect. Nonmatched features will vary from foil to foil.

Admonishment strength will be controlled within subjects and counterbalanced across the design of the study. Participants given the strong admonishment condition first will encounter the following: A very serious experimenter wearing a lab coat will deliver the instructions and the laboratory environment will be made to look very orderly. Participants will be verbally instructed that the culprit may or may not be present in the lineup. They will also be further informed that the results from the study will be used to guide the actual practice of the legal system. As such, they should perform their very best. Furthermore, similar to real world witnesses, they should keep in mind the seriousness of the task. They should try to act like real world witnesses, and real world witnesses know that wrongfully accusing an innocent person has devastating consequences. After completing phase 1, participants will be given a break. In phase 2, the weak admonishment condition will be run. When subjects return to the laboratory, they will overhear a new, less serious minded experimenter complaining that the laboratory principal investigator never publishes any of their experiments. The new experimenter will then run phase 2. Participants will simply be told that the culprit may or may not be present in the lineup. The new experimenter will also have their personal belongings strewn about the room.

A total of 192 subjects will be run in the experiment. Power to detect a difference in correct identifications between simultaneous and sequential lineups with a small effect size (d = .25) is .76.

IV. Procedural Issues: Criterion Placement and Witness Choice Rates

The number of pictures that witnesses expect to see and how readily witnesses can gauge task difficulty are two variables that further differentiate simultaneous and sequential lineup presentations. In the previous experiment, subjects viewing sequential lineups will know that they are going to see only 6 pictures. In this experiment, subjects will be led to think that up to 20 pictures will be shown. Furthermore, in simultaneous presentations, witnesses can immediately determine the difficulty of the identification task. Witnesses viewing sequential lineups, however, will not know until after several pictures have been shown. Therefore, subjects will also be given a practice lineup, composed of members that are either highly similar or dissimilar to the target to determine whether criterion placement is affected by perception of task difficulty. A final criterion based manipulation that will be attempted is to instruct participants to use either a holistic strategy (i.e., compare the personalities of the lineup members) or a featural strategy (compare the features of the lineup members).

Participants will view a video of a masked armed robber. A composite drawing of the culprit will appear on screen for 10 seconds and the identification procedure will be conducted 4 minutes later. Lineup presentation method will be controlled between subjects.

Pictorial stimuli will be drawn from the previous experiment. The participant will view either a target present or a target absent lineup. The similarity of the lineup foils relative to the suspect will be controlled between subjects (3 features or 1 feature is matched). The innocent suspect will always have 3 features in common with the perpetrator.

A total of 256 subjects will be run in this experiment.

V. Constructing Lineups Matching Suspect's Description or Photo

This experiment will determine whether findings from the previous experiments generalize to photographic lineups. Photos of college students and actual criminal offenders (selected from police databases located on the Internet) will be used as the stimuli. Distinctive persons (i.e., those with tattoos, scars, missing teeth, and eye patches) will be excluded from the stimulus pool. (Though the rate at which distinctive photos are sampled will be recorded.) Subjects assigned to play the role of "police investigator" will create lineups. They will be instructed to match either the photo or the description of the target using a predetermined number and type (hair, eyes/eyebrows, and face shape) of features. Other facial features that are not being matched in a given condition will be covered. The pool of available foils from which lineup members are drawn will match the modal description of the culprit given by an independent group of subjects.

Investigators will make a total of 8 lineups, one per suspect. The feature(s) that they will be asked to match for a given suspect will be randomly determined, with the constraint that they make one lineup for every possible feature combination. For the remaining suspect, the lineup will be made without any instructions other than to match the suspect’s photo or description. Investigators will also select an innocent suspect for the perpetrator by finding a picture that matches the hair, eyes/eyebrows, and face shape of the culprit.

Subjects will attempt to identify a total of 16 suspects, with 8 suspects presented per phase of the experiment. The suspect will be presented for 10 seconds and 4 minutes later subjects will view a target present and a target absent lineup (order of presentation will be random). Lineup presentation method (simultaneous or sequential) will be switched between the 2 phases of the experiment. In each phase, half of the suspects will be college students, and the other half will be criminal offenders.

References

Behrman, B. W., & Davey, S. L. (2001). Eyewitness identification in actual criminal cases: An archival analysis.

Law & Human Behavior, 25(5), 475-491.

Clark, S. E. . WITNESS: A mathematical model of eyewitness identification. Paper presented at the Society for Applied Research in Memory and Cognition, Boulder, CO.

Cottrell, G. W., Dailey, M. N., Padgett, C., & Adolphs, R. (2001). Is all face processing holistic? The view from UCSD. In E. Michael J. Wenger & E. James T. Townsend (Eds.), Computational, geometric, and process perspectives on facial cognition: Contexts and challenges. (pp. ix, 514): Lawrence Erlbaum Associates, Inc., Publishers.

Cutler, B. L., & Penrod, S. D. (1988). Improving the reliability of eyewitness identification: Lineup construction and presentation. Journal of Applied Psychology, 73(2), 281-290.

Ebbesen, E. B., & Flowe, H. F. (2001). Simultaneous v. Sequential Lineups: What Do We Really Know? Unpublished manuscript.

Flowe, H. D., Ebbesen, E. B., Burke, C., & Chivabundtt, P. (2001). At the scene of the crime: An examination of the external validity of published studies on line-up identification accuracy. Paper presented at the American Psychological Society, Toronto, Canada.

Flowe, H. F., & Ebbesen, E. B. (2001). Mistakenly identified: Measuring bias in a police and laboratory photographic lineups. Paper presented at the Society for Applied Research in Memory and Cognition, Toronto, Canada.

Flowe, H. F., & Ebbesen, E. B. (2002). An Archival Analysis of Trial Cases in which the Defense Claimed Mistaken Identification. Paper presented at the American Psychology-Law Society, Austin, TX.

Haig, N. D. (1984). The effect of feature displacement on face recognition. Perception, 13(5), 505-512.

Levi, A. M. (1998). Protecting the innocent, nailing the guilty: A modified sequential lineup. Applied Cognitive Psychology, 12, 265-275.

Lindsay, R. C., & Wells, G. L. (1985). Improving eyewitness identifications from lineups: Simultaneous versus sequential lineup presentation. Journal of Applied Psychology, 70(3), 556-564.

Lindsay, R. C., Wallbridge, H., & Drennan, D. (1987). Do the clothes make the man? An exploration of the effect of lineup attire on eyewitness identification accuracy. Canadian Journal of Behavioural Science, 19(4), 463-478.

Lindsay, R. C., Lea, J. A., Nosworthy, G. J., & Fulford, J. A. (1991). Biased lineups: Sequential presentation reduces the problem. Journal of Applied Psychology, 76(6), 796-802.

Lindsay, R. C. L., & Bellinger, K. (1999). Alternatives to the sequential lineup: The importance of controlling the pictures. Journal of Applied Psychology, 84(3), 315-321.

Luus, C. E., & Wells, G. L. (1991). Eyewitness identification and the selection of distracters for lineups. Law & Human Behavior, 15(1), 43-57.

Navon, D. (1992). Selection of lineup foils by similarity to the suspect is likely to misfire. Law & Human Behavior, 16(5), 575-593.

Parker, J. F., & Ryan, V. (1993). An attempt to reduce guessing behavior in children's and adults' eyewitness identifications. Law and Human Behavior, 17(1), 11-26.

Rhodes, G., Brake, S., Taylor, K., & Tan, S. (1989). Expertise and configural coding in face recognition. British Journal of Psychology, 80(3), 313-331.

Sporer, S. L. (1993). Eyewitness identification accuracy, confidence, and decision times in simultaneous and sequential lineups. Journal of Applied Psychology, 78(1), 22-33.

Steblay, N. M. (1997). Social influence in eyewitness recall: A meta-analytic review of lineup instruction effects. Law & Human Behavior, 21(3), 283-297.

Tanaka, J. W., & Farah, M. J. (1993). Parts and wholes in face recognition. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 46(2), 225-245.

Wall, P. (1965). Eyewitness Identification in Criminal Cases. Springfield: Thomas.

Wells, G. L. (1984). The psychology of lineup identifications. Journal of Applied Social Psychology, 14(2), 89-103.

Wells, G. L., Small, M., Penrod, S., Malpass, R. S., Fulero, S. M., & Brimacombe, C. A. E. (1998). Eyewitness identification procedures: Recommendations for lineups and photospreads. Law & Human Behavior, 22(6), 603-647.

Williams, C. (1955). Identification Parades. Criminal Law Review, 525-537.

Wogalter, M. S., Marwitz, D. B., & Leonard, D. C. (1992). Suggestiveness in photospread line-ups: Similarity induces distinctiveness. Applied Cognitive Psychology, 6(5), 443-453.

Wogalter, M. S., Burger, M. A., & Malpass, R. S. (1993). Methods of constructing live and photographic lineups by police officers in the United States. Paper presented at the International Conference on Face Processing, Wales, United Kingdom.

Figure Captions

Figure 1. Wells (1993) removal without replacement procedure results.