Download Person Descriptions as Eyewitness Evidence and more Lecture notes Cognitive Psychology in PDF only on Docsity!
Person Descriptions
as Eyewitness Evidence
Christian A. Meissner
University of Texas at El Paso
Siegfried L. Sporer
University of Giessen, Germany
Jonathan W. Schooler
University of British Columbia, Canada
Two teenage girls were enjoying their family vacation in a hotel hot tub one evening. Shortly after their parents had left them, the girls were approached by a stranger, who proceeded to join them in the hot tub. Following several minutes of conversation, the stranger attempted to molest the older girl by touching her “private parts.” The older girl struck the stranger in the face and told him to stop, and instructed the younger girl to find their parents. After the older girl shouted “rape” several times, the stranger finally exited the hot tub, gathered his belongings, and ran from the area. The girls would later describe the stranger as a male in his twenties, with no shirt, wearing tan/brown shorts and a shell necklace. He had dark hair, and a dark/suntanned complexion. Upon receiv- ing the description, detectives released a BOLO (“be on the lookout”), and a suspect matching the description was detained 45 minutes later as he walked on the beach about a half-mile from the hotel. The girls were brought to the suspect and together identified him as the stranger they had encountered. The suspect was arrested for the crime, but prosecutors would later drop the charges when the suspect provided a detailed (and cor- roborated) alibi for his whereabouts at the time of the incident. Simply put, detectives had detained the wrong person. Person descriptions represent an important element for detectives in the investiga- tion of any crime. Unfortunately, the descriptions provided by witnesses or victims tend to be rather nondistinct and, like the description provided by the teenage girls above, can frequently apply to many people in the vicinity of the crime. Although descriptions
3
4 MEISSNER, SPORER, SCHOOLER
are most often useful for locating a suspect in the immediate aftermath of an incident, they are also used throughout a criminal investigation to identify potential suspects from mug books, to construct sketches or composites of a suspect, and as a basis for selecting fillers when investigators are constructing a lineup identification parade and subsequently assess the “fairness” of that lineup. In addition, witness descriptions are regularly intro- duced at trial as a means for demonstrating the congruence between the suspect and a witness’s memory. In Neil v. Biggers (1972), the U.S. Supreme Court indicated that wit- ness descriptions could be used as one basis for determining the accuracy of a witness. However, as discussed in this chapter, the relationship between a witness’s description and his or her ability to perceptually identify the actual perpetrator is not clear-cut. Given the importance of person descriptions as eyewitness evidence, psychologists and criminologists have conducted a wealth of research aimed at establishing what is known about the content and veracity of person descriptions, as well as factors that may positively or negatively influence a witness’s ability to provide an accurate description. The current chapter provides a review of this research, including a discussion of psycho- logical factors that may influence person descriptions at encoding (e.g., alcohol, stress, illumination, distance, etc.), the effects of delay and repeated descriptions over time, the role of person variables (e.g., age, gender, race, etc.) and individual differences, and the influence of misinformation from investigators and/or co-witnesses. In addition, we ad- dress the variety of recall techniques that have been explored to improve the quality and quantity of person descriptors, and the relationship between such description procedures and witnesses’ subsequent attempts at perceptual identification of a suspect (i.e., the verbal overshadowing effect, Schooler & Engstler-Schooler, 1990; or the use of person descriptions as retrieval cues, Cutler, Penrod, & Martens, 1987; Sporer, in press).
CONTENTS OF PERSON DESCRIPTIONS
IN REAL CASES
Quantity and Quality of Descriptors
A number of archival studies have examined the content of person descriptions in real cases. Likely the most well-known study was conducted by Kuehn (1974). This study involved the analysis of person descriptions contained in 100 police protocols of cases of bodily injury, rape, and robbery in Seattle, Washington. Statements were taken from the witnesses immediately after the incident, and all perpetrators in the sample were strangers. Unfortunately, it is not clear from Kuehn’s report whether the descriptions were rendered as free descriptions or were the result of some standardized questioning scheme employed by the local police. The number of details contained in the descrip- tions was fairly meager overall (with a maximum of nine descriptors)—on average there were 7.2 descriptors, whereas most witnesses reported 8 or 9 features. Only four vic- tims were unable to provide any details at all. In descending order of frequency, gender, age, height, build, race, weight, complexion, and hair color were mentioned. With the exception of eye color (23%), all features were named by more than 70% of all vic-
tioned jewelry or the dialect spoken. Close to 5% of the descriptors referenced personal- ity characteristics (which are useless when investigators are trying to find a person to ar- rest, but may promote subsequent recognition of the person because of the deeper level of processing possibly involved at encoding; see Sporer, 1991). It is also noteworthy that quantity and pattern of descriptions found in this archival study closely resembled those of a staged event study in which a confederate had interrupted a lecture to take away a slide projector (Sporer, 1992b). Of the facial descriptors analyzed by Sporer (1992a), the majority referred to the upper half of the face, particularly the hair of the perpetrator. This finding confirms ear- lier studies on contents of facial descriptions (Ellis, Shepherd, & Davies, 1980; Laughery, Duval, & Wogalter, 1986; Shepherd, Ellis, & Davies, 1977) and on the importance of upper portions of the face in the recognition process (e.g., Fisher & Cox, 1975). In fact, the cheek and chin (but also the forehead) were rarely mentioned in these descriptions. Although reference to hair (about 16% of all descriptors) appeared to be the most dom- inant single descriptor, it is also the most problematic and is likely to be of little help in the pursuit of a criminal because hair style can be most readily altered in comparison with other more permanent features (e.g., inner features of a face). Similarly, the large number of references to the clothing of the perpetrator is generally of limited value when police are attempting to locate a perpetrator. Lindsay, Martin, and Webber (1994) examined the descriptions of 105 criminals pub- lished in the Kingston, Ontario newspaper ( The Whig Standard ) and compared their com- pleteness with that of 100 descriptions (across five targets) obtained from a series of lab- oratory studies. Participant-witnesses viewing staged crimes were most likely to report clothing (99%), hair color (90%), and height (86%), whereas less than 50% reported such obvious descriptors as gender, age, or race/ethnicity. The most frequently reported feature of the face was the eyes (43%), and all other features were reported less than 25% of the time. Witnesses to real crimes were significantly more likely to report gender (96%), hair color (38%), clothing (60%), and race/ethnicity (25%), and facial features were provided in less than 10% of the sample. Although the results of Lindsay et al. indicated that laboratory witnesses provided more complete descriptions than real witnesses (7. vs. 3.94 features, respectively), they more generally concluded: “The data strongly sup- port our concern that eyewitness descriptions are frequently vague” (p. 531). Van Koppen and Lochun (1997) reported a large-scale archival analysis of person descriptions in 431 robbery cases. A total of 1313 witnesses provided 2299 descriptions of the offenders. Descriptors were subdivided into 24 permanent descriptors (e.g., gen- der, skin color) and 19 temporary characteristics (e.g., particulars of clothing, type of mask). Similar to Sporer’s (1992a) findings, the completeness of the descriptions was rather poor. Of the possible maximum of 43 descriptors, the median number provided by each witness was 8 (interquartile range 6). Permanent features were mentioned more frequently (median 5, interquartile range 5) than temporary characteristics (median 2, interquartile range 3). Considering that gender, appearance (including race), and skin color were among the most frequently mentioned permanent character- istics (characteristics that are likely the most obvious to any observer), the paucity of these descriptions becomes even more dramatic. Less than 5% of the descriptors referred
6 MEISSNER, SPORER, SCHOOLER
to inner features of a face (eye color, nose, face color or complexion, mouth, eye shape, teeth, earrings, chin, ear size, ears protruding), which are considered most important for identifying another person (Ellis, 1992). Of the temporary characteristics, the majority of descriptors referred to hats (51%) and hat color (31%), as well as jackets (28%), coats (25%), and trousers (26%), and their respective colors (28%, 22%, and 18%). Van Koppen and Lochun’s (1997) analysis was not restricted to the quantity of in- formation recalled as in Sporer’s (1992a) study, but also sought to analyze the accuracy of descriptions by validating the descriptions by witnesses against the descriptions con- tained in the police database used in the Netherlands. Although more elements of the descrptors were correct than incorrect, the majority of crucial facial descriptors were wrong (e.g., accuracy of facial descriptors included: eye color 36%; nose 35%; mouth 39%; chin 38%). Most strikingly, almost all descriptors of facial hair (beard and mustache) failed to match the police database. Given that perpetrators may have changed these aspects of their appearance over time, however, the latter finding is diffi- cult to interpret. Interestingly, there was a negative correlation between accuracy and completeness, indicating that when witnesses did provide more extensive descriptions their accuracy suffered.
Estimates of Height and Weight
Almost all person descriptions contain references to the perceived height, weight, and age of the perpetrator (Kuehn, 1974; Sporer, 1992a; van Koppen & Lochun, 1997; Yuille & Cutshall, 1986); however, authors differ in their interpretation of existing data re- garding the extent to which such estimates are accurate. Some authors have defined ac- curacy as estimates falling within a certain range of “true” values (e.g., true value plus or minus 2 inches or 5 pounds; see Yarmey & Yarmey, 1997; Yuille & Cutshall, 1986), con- cluding that estimates appear to be rather accurate. Then again, treating values with a difference of 4 inches (almost 10 cm) in height as “accurate” would allow an estimate of 170 cm to be equivalent to one of 180 cm, values that are substantially below or above the population average (see Flin & Shepherd, 1986; Sporer, 1996). When the accuracy of estimates for height and weight are defined as the correlation between the actual values and their estimates, these correlations are well below their maximum possible value. For example, Janssen and Horowski (1980) reported that the average correlations between the actual and estimated heights in a series of studies with students aged 10 to 18 fluctuated between .26 r s .90. As might be expected, the correlations were smaller for younger children than for older teenagers. This age effect could be either a function of the restricted experience of the smaller children with num- bers (see also Davies, 1996) or a result of the smaller children’s own height, which seems to assist adults in gauging their estimates of another person. Next to the target’s true height and weight, probably the most important determi- nant of this type of estimate is the witness’s own height and weight, perhaps modified by his or her knowledge (or better, supposition) of what the average population norm might be for a typical middle-aged male or female. Flin and Shepherd (1986) have presented a comprehensive and representative study on this topic. The authors had 588 participants
- PERSON DESCRIPTIONS 7
Although it has been assumed that the opportunity to view a target person (i.e., dis- tance between or duration of the event) should significantly influence the accuracy or completeness of person descriptions, only a handful of field studies have attempted to in- vestigate such factors. For example, Yarmey, Jacob, and Porter (2002) conducted a study in which participants interacted with a target person for 5 seconds or 30 seconds and were subsequently asked to describe the encounter. As expected, their results indicated that person descriptions (particularly for clothing) were superior when participants had a longer time to observe the target person. Another aspect that appears to be important regards whether the witness encodes information about a perpetrator with the intent of later recalling it from memory. Along these lines, Yarmey (2004) found that instructions to intentionally encode information from the event for a subsequent memory test led to superior recall of person descriptions (again particularly for articles of clothing). Although both laboratory and field research on such factors has been minimal, there are some archival analyses of criminal records that have explored the importance of viewing conditions. Despite claims for the superior ecological validity of archival studies (Yuille & Cutshall, 1986), the problem with archival analyses is that the accuracy of the descriptions generally cannot be determined—rather, a proxy for accuracy must be cre- ated with respect to the precision of the description, or its relative consistency with that of the individual found guilty for the crime. In Sporer’s (1992a) study, the mean number of descriptors, length, and precision of person descriptions were coded and related to low, medium, and high levels of a host of potentially relevant factors, including illumi- nation, duration of event, and time to observe. The categories “low,” “medium,” and “high” are not to be taken literally, as they may take on different meanings with respect to the particular variable coded (e.g., “high illumination” was operationalized as bright daylight or good artificial lighting). Level of illumination had the expected effect (such that greater illumination led to more complete person descriptions), whereas duration of the incident and time estimated for the target to be in view did not seem to influence description completeness. Similarly, van Koppen and Lochun (1997) found that better illumination and shorter distances between the witness and perpetrator were associated with greater frequency of person descriptors. Whereas both of these studies supported the predicted linear relationship between opportunity to view and recall, Kuehn’s (1974) archival analysis found worse performance for twilight conditions than for observations either at bright daylight or at night.
Stress or Anxiety. Eyewitness events are generally considered anxiety-provoking situations in which the victim or witness is likely to experience a great deal of stress during the encoding process. Consistent with this notion, a number of studies of eye- witnesses have suggested that high levels of stress or anxiety impair memory by restrict- ing attentional and executive processes at encoding and thereby prevent the consolida- tion of information into a coherent event sequence (see Deffenbacher, 1983, 1994). On the other hand, other studies suggest that stress may increase participants’ memory for central details (Christianson, 1992) and that the negative effects of stress (at least in some cases) may reverse with the passage of time (Burke, Heuer, & Reisberg, 1992; Christian- son, 1984; for a general review see Deffenbacher, Bornstein, Penrod, & McGorty, 2004;
- PERSON DESCRIPTIONS 9
Schooler & Eich, 2000.) With regard to person descriptions, several laboratory studies have demonstrated impairment in accuracy and completeness as a result of stress or anx- iety. For example, Clifford and Hollin (1981) varied the violence of a to-be-remembered event and found that participants in the violent conditions were less likely to recall de- tails of the perpetrator (see also Loftus & Burns, 1982). In their recent meta-analysis on the topic, Deffenbacher et al. (20004) found that heightened anxiety led to significant decrements in recall accuracy (Cohen’s d .31) across studies. The presence of a weapon, which may be accompanied by stress or fear, has also been shown to divert a witness’s attention away from the face of the offender. A number of studies have investigated the possibility that the presence of a weapon is associated with impaired recall of details of the perpetrator or event. Consistent with the afore- mentioned research, studies of the “weapon focus” effect have generally demonstrated a significant influence of the presence of a weapon on person description accuracy (see meta-analysis by Steblay, 1992). Recent research by Pickel (1998, 1999) has indicated that the unusual or unexpected nature of a weapon may be responsible for the observed effect on description accuracy, when contrasted with the “threat” posed by the object. Archival studies of eyewitness testimony have also attempted to assess the influence of anxiety, stress, or the presence of a weapon on the accuracy or completeness of person descriptions. Given that stress in criminal situations could not be observed (or manipu- lated) directly, the amounts of anxiety and arousal were coded retrospectively by classi- fication of an event on the basis of the reports emerging from police records (e.g., pres- ence of a deadly weapon, bodily injury, etc.) or of self-reports of anxiety provided by the witnesses in the course of testimony. In Sporer’s (1992a) study, three groups of witnesses were compared: victims, bystanders participating in the event without being victims, and other witnesses who were questioned by the police about the perpetrator during the investigation but were not themselves directly involved in the case (e.g., the owner of a gunshop where the perpetrator bought his weapon). Overall, the most striking finding of this analysis was that none of the various ways in which stress had been coded seemed to indicate the expected deterioration in witness recall for high levels of stress and its asso- ciated variables. In fact, there even appeared to be a (linear) increase in descriptive de- tails as a function of some of these stress-related variables (e.g., greater reported anxiety was associated with a greater number of details). An analysis of stress conducted by Yuille and Cutshall (1986) showed similar results, whereas an analysis conducted by van Koppen and Lochun (1997) demonstrated results consistent with the laboratory and field research reported earlier (i.e., high levels of stress associated with impaired recall performance). A more recent archival study by Wagstaff et al. (2003) demonstrated null effects on the accuracy or completeness of person descriptions. The general inconsistency observed between laboratory or field research and archival research may potentially be accounted for by length of the retention interval. Laboratory studies have typically used short retention intervals that are known to sometimes give an advantage to nonstressful memories, whereas archival studies typically involve longer retention intervals, which sometimes afford advantages to more stressful memories (Kleinsmith & Kaplan, 1963, 1964). It also possible that stressful experiences may be more likely to incur rehearsal, which could increase the amount of details recalled. Importantly, in none of these
10 MEISSNER, SPORER, SCHOOLER
another either 1 hour, the next day, or 1 week following exposure. Participants remem- bered significantly fewer details after 1 week compared with the two shorter retention intervals, and memory loss was rather equally distributed across specific facial features. The accuracy of person descriptions also declined significantly with the longer delay in- terval. In a similar laboratory experiment, Meissner (2002) found significant losses in both the completeness and the accuracy of facial descriptors when participants provided a description either immediately or following a 1-week delay. In their archival analysis, van Koppen and Lochun (1997) observed a pattern con- sistent with the aforementioned laboratory studies, such that witnesses provided fewer person descriptors following longer retention intervals. In contrast to this study, Yuille and Cutshall (1986) and Cutshall and Yuille (1989) emphasized strikingly high levels of recall from witnesses of real crimes as late as 2 years after the incidents. As mentioned previously, these high levels of performance were likely mediated by repeated question- ing (and rehearsal) prior to recall at the time of the study (see Sporer, 1989). It should be noted that the course of time alone is unlikely to have a detrimental effect on recall; rather, both the strength of the initial memory trace and interference from a variety of activities during the delay interval are likely the major influence of a witness’s ultimate recall of person descriptors. Generally referred to as “post-event infor- mation,” witnesses may obtain information during the retention interval (either deliber- ately or unintentionally) through a number of sources or tasks that they engage in. For example, overhearing a description provided by another person or being shown an erro- neous facial composite or sketch can lead the witness to incorporate erroneous details into his or her own description of the perpetrator, and the likelihood of such post-event information influencing subsequent recall has been shown to increase following a long retention interval (Loftus & Greene, 1980; Loftus & Ketcham, 1983; Shaw, Garven, & Wood, 1997; Sporer, 1996b). The related effects of misleading questioning by investiga- tors (referred to as “misinformation”) and collaborative recall with another witness (or “co-witness” effects) are discussed below.
Witness and Target-Person Variables
As in the eyewitness identification literature, a number of witness and target variables (e.g., gender, age, ethnicity, etc.) appear to influence the accuracy and completeness of person descriptions. This section reviews the available literature on such variables.
Gender. Although many studies on eyewitness memory have included both male and female participants, few have analyzed gender differences. Several studies conducted by Yarmey (1986, 1993, 2004) have generally indicated few differences in the recall of men and women. When differences were noted, they typically involved responses to spe- cific attributes that women may have been more likely to attend to at encoding (e.g., jewelry, hair color or length, and weight; see Yarmey, 2004), or they involved more com- plex interactions between variables (such as levels of illumination; see Yarmey, 1986). In several studies, Yarmey noted that men appeared more confident in their responses than women (Yarmey, 1986, 1993).
12 MEISSNER, SPORER, SCHOOLER
MacLeod and Shepherd (1986) have drawn attention to gender differences in an archival study of criminal assault cases. Similar to research by Yarmey (1986, 2004), gen- der differences were found to covary in a complex manner with such variables as the type of questions analyzed (e.g., action details vs. descriptive details; statements refer- ring to self, victim, accused, or periphery) and the type of incident (involving injury of the victim or not). In his archival analysis, Sporer (1992a) reported that male witnesses provided on average longer descriptions than females ( M 7.50 vs. 7.10 number of lines in the protocol, respectively). In contrast, the number of descriptors and rated pre- cision of statements showed an opposite but nonsignificant trend favoring females. Thus, it appears that although females may have said less quantitatively, they did not necessarily convey less information.
Child Witnesses. Although some studies have found that the relative accuracy in reports of children may not differ from that of adults (Goodman & Reed, 1986; Leippe, Romanczyk, & Manion, 1991; Marin, Holmes, Guth, & Kovac, 1979), adults’ state- ments are likely to be much longer and more detailed than those of children (Davies, Tarrant, & Flin, 1989; Dent & Stephenson, 1979; Leippe et al., 1979; Marin et al., 1979). In contrast, a recent study conducted by Pozzulo and Warren (2003, Experiment 1) observed both greater accuracy and completeness of person descriptions provided by adults versus youths (ages 10 to 14). Further analyses indicated that adults were more likely to report features of the face, aspects of the body (i.e., height, weight, and build), and race of the perpetrator, whereas youths were more likely to report various acces- sories (e.g., belt or glasses). With regard to accuracy, youths were significantly less accu- rate than adults in describing interior facial features (e.g., eyes, nose, or mouth), aspects of the body, and the age of the perpetrator. In a follow-up study using a live event, Poz- zulo and Warren (Experiment 2) observed the more classic pattern involving a greater frequency of person descriptors by adults when compared with youths, but no differ- ences in the overall accuracy of features reported. The analysis of specific features was largely consistent with the first study, except that aspects of the target’s clothing were more likely to be reported by adults in the sample. Recent research by Lindholm (2005) has also suggested that witnesses, particularly children and young adults, may actually perform better when recalling descriptions of target persons matching their own age group. Such own-age effects (similar to the cross-race effects discussed below) may result from a variety of experiential or motivational factors (cf. Sporer, 2001a), and further research on this topic seems warranted. Saywitz (1995) has suggested that it may be important to adapt one’s language when interviewing children such that questions are more comprehensible to young children. In particular, interviewers should use short sentences with a simple grammatical struc- ture, common phrases, and proper names. They should avoid the passive voice, double negatives, and indirect questions. Before estimates are obtained, interviewers should also make sure that children understand concepts like size, distance, weight, age, and time, as well as particular body parts and various color names. For example, Dent (1982) reported large inaccuracies in estimates with children between 8 and 13 years of age. Furthermore, age estimates may suffer from children’s lack of knowledge of facial cues to
- PERSON DESCRIPTIONS 13
and accuracy for young versus elderly adults. However, to the extent that most of these studies have used only young adults as targets to be observed and described, these stud- ies may reflect as much an in-group bias in the form of an own-age effect (Sporer, 2001a) as deficits in the memory of elderly witnesses.
Cross-ethnic Differences. Although more than 60 studies have investigated recog- nition memory for own- versus other-race faces (for reviews see Chance & Goldstein, 1996; Meissner & Brigham, 2001b; Sporer, 2001a), very few studies have attempted to determine whether participants differ in the way they describe faces of their own and another race (Sporer, 2001b). Those that have investigated descriptions of own- versus other-race faces have suggested that individuals attend to features deemed relevant to own-race faces and further attempt to apply this encoding scheme inappropriately when examining other-race faces (Ellis, Deregowski, & Shepherd, 1975; Shepherd & Dere- gowski, 1981). For example, Ellis and colleagues (1975) demonstrated several differ- ences in the type of features that black and white participants recalled (regardless of the race of face). Although Ellis and colleagues did not assess descriptions for accuracy or discriminability, they did note that white participants often reported rather “redundant” descriptions of black faces (e.g., “he has black skin, black, kinky hair and brown eyes”) that would likely be indiscriminant upon later assessment (p. 123). Fallshore and Schooler (1995) compared Caucasian undergraduates’ ability to iden- tify and describe African American and Caucasian faces. As is typically found, they observed the cross-race effect for lineup identification decisions, such that participants were better able to recognize Caucasian relative to African American faces. However, when description accuracy was assessed with the use of a communication accuracy par- adigm in which subject-judges attempted to identify the faces based on witnesses’ verbal descriptions, no cross-race effect was observed (although a numerical advantage was shown for the identification of other-race faces). Fallshore and Schooler speculated that differences in the pattern of results associated with cross-racial face recognition versus face description may be due to differential reliance on configural versus featural process- ing for own versus other race faces, respectively (see Rhodes, Brake, Taylor, & Tan, 1989). Accordingly, if the source of the own-race face recognition advantage were an enhanced ability to rely on configural information (Sporer, 2001a), then it follows that verbal de- scription ability, which typically relies on featural knowledge (see Farah, Wilson, Drain, & Tanaka, 1998; Wells & Turtle, 1987), should not reveal such differences. Thus, al- though the relative dearth of studies on the topic clearly suggests the need for additional research, the absence of evidence for an own-race advantage for person description may reflect fundamental differences in the processes associated with face recognition versus description.
Methods for Obtaining Person Descriptions
Several methods of eliciting a person description have been developed over the years, from standard free recall approaches to feature checklists and techniques based upon
- PERSON DESCRIPTIONS 15
principles of cognitive psychology (e.g., the cognitive interview). In this section, we dis- cuss research on the generation of person descriptions and their positive and negative effects. Along the way, we also address the role of leading questions and attempts at per- mitting witnesses to collaborate in generating a description, and we consider the effect of repeated questioning on the accuracy and completeness of person descriptions.
Free Recall vs. Leading Questions. Likely the most common technique used by investigators to obtain a person description involves a request for the witness to simply recall what he or she remembers about the perpetrator of the crime. Although such free recall descriptions are often quite accurate, unfortunately they rarely satisfy investigators, because of their likelihood of being incomplete with regard to critical details (Lipton, 1977). Thus, investigators will frequently follow up with more specific, close-ended ques- tions to complete the description (e.g., Do you remember the color of the man’s hair? ). In addition, investigators may have previously received information regarding the perpetra- tor and so will attempt to confirm this information by inquiring about more specific de- tails (e.g., Did the man have red hair with long sideburns? ) or may include this information in the context of inquiring about another detail (e.g., This man with the red hair and long sideburns, did he have any facial hair? ). Unfortunately, such leading questions can have rather harmful consequences for the witness’s attempts at subsequent recall, as studies indicate that witnesses are quite likely to incorporate potentially inaccurate information (“misin- formation”) into their person descriptions (Loftus, 1975, 1979; Loftus & Zanni, 1975). For example, Loftus and Greene (1980) observed that participants who viewed a face and then heard a description of the face that was attributed to another witness later in- corporated the verbal expressions of that witness into their description, even when the description was in error.
Feature Checklists. As noted above, one primary drawback to the use of free recall tasks regards the incompleteness of person descriptions. Witnesses will often vary in their output criterion for recalling details of an event (Koriat & Goldsmith, 1996), and a com- mon difficulty with person descriptions involves the limited vocabulary that individuals have for describing the human face. In an attempt to alleviate this problem, researchers have sought to develop feature checklists that might aid witnesses in providing more complete (and useful) descriptions of the perpetrator they viewed. For example, Shep- herd (1986; see also Shepherd & Ellis, 1996; Sporer, in press) and his colleagues have developed the Aberdeen Face Rating Schedule, which consists of some 50 items on which witnesses are asked to rate individual features of a face on five-point scales (for a published version of these scales, see Sporer, in press). Using these forms, observers are prompted to use certain features that otherwise they might omit or forget. However, ac- curacy of these descriptions might be poor, as people may frequently mark the middle (“normal”) value of the scale when they either don’t remember or guess the information (Sporer, in press). Nonetheless, forms of this type are useful both for communicating in- formation to other agencies and for conducting computerized searches to identify indi- viduals in mug shot databases who might be presented to the witness (cf. Pryke, Lindsay, & Pozzulo, 2000). A prototype of such a system was developed by psychologists at the
16 MEISSNER, SPORER, SCHOOLER
Repeated Questioning. Witnesses may be asked to provide a description of the perpetrator and event on multiple occasions, including immediately following the event, throughout the investigative process, in depositions and pretrial hearings, and finally (but most importantly) on the witness stand before a jury. To what extent might repeated questioning influence the veridicality of the information provided by the witness? The general cognitive literature has shown both positive and negative effects of repeated re- call (Brown, 1923). For example, individuals may benefit from repeated attempts by recalling information or items that had not previously been reported (Payne, 1987; Roediger & Challis, 1989). To avoid confusion, we adopt the distinction between hyper- mnesia , that is, an increase in net recall (number of new details minus number of items lost), and reminiscence , that is, the gross recall of details provided at least once across a number of trials (Payne, 1987; Turtle & Yuille, 1994). In their study of eyewitnesses, Scrivner and Safer (1988) demonstrated hypermnesia during the repeated recall of event and perpetrator details from a previously viewed crime. Turtle and Yuille (1994) par- tially replicated these findings with longer retention intervals between successive re- call episodes, demonstrating reminiscence but not hypermnesia. Bornstein, Liebel, and Scarberry (1998) further demonstrated that repeated testing can improve recall for de- tails of a negatively arousing event. In addition to the possibility of more complete descriptions, repeated testing has also been shown to preserve an individual’s memory by strengthening associations that are retrieved (see Bjork, 1988). One important moderator, however, regards the reten- tion interval prior to the first attempt at retrieval—to the extent that the retention in- terval is brief, more information may be preserved by the act of retrieval (Bahrick, 2000; Ebbesen & Rienick, 1998; Shaw, Bjork, & Handall, 1995). Ebbesen and Rienick (1998) varied the interval between exposure to a target and the first recall attempt (1 day, 7 days, or 28 days), and all participants provided a second recall attempt after 4 weeks. Their results indicated that, across all conditions, participants recalled about one less descriptor at the 4-week test ( M 8.50) than at all other tests ( M 9.50). Although the authors stress the fact that there was virtually no decline in the recall of personal attributes once a recall attempt was made, the percentage of errors for facial features, clothing color, and clothing style was still substantial. Even recall for the ethnicity of the person who participants had interacted with showed error rates between 13% and 23%. Nonetheless, these results do appear to demonstrate the predicted protection of person description memory afforded by repeated questioning. In a similar fashion, Dunning and Stern (1992) reported two experiments in which participants showed a (nonsignificant) tendency to recall more person information correctly, with no change in incorrect or confabulated details, over repeated reports. The interval between reports, however, was only 5 minutes, which is functionally quite different from the situation in which wit- nesses are repeatedly asked about events at different occasions separated by days or even months (Sporer, 1992a; van Koppen & Lochun, 1997). In contrast to the benefits of increased completeness and maintenance of the mem- ory, Roediger and his colleagues have demonstrated that repeated testing can also have rather paradoxical effects in which erroneous information may be reported and incor- porated into subsequent recall episodes (see Roediger, McDermott, & Goff, 1997; Roedi- ger, Wheeler, & Rajaram, 1993). For example, a study by Roediger, Jacoby, and McDer-
18 MEISSNER, SPORER, SCHOOLER
mott (1996) demonstrated that when participants were encouraged to recall erroneous information from a previously viewed crime, they were more likely to report that infor- mation in later attempts at recall (cf. Schooler, Foster, & Loftus, 1987). Meissner (2002) subsequently replicated the pervasive effects of self-generated misinformation in the context of person descriptions, particularly when participants were forced to report descriptors that they were unsure of.
Cognitive Interview. Over the years, researchers have been interested in devising techniques that might improve the accuracy and completeness of information obtained from witnesses. Likely the most well-known technique is the cognitive interview, which was initially developed by Geiselman and Fisher in the early 1980s (for a review, see Fisher & Geiselman, 1992). Overall, the cognitive interview consists of four main com- ponents: (1) context reinstatement, which includes mentally reinstating the environ- mental and personal context of the original event; (2) instruction to “report all” infor- mation, including partial information, even if it seems unimportant; (3) recounting the event in a variety of temporal orders; and (4) reporting the events from a variety of per- spectives. With the use of the cognitive interview, a host of studies have shown that de- scriptions of persons, objects, and events can be reliably improved when compared with other standard (free recall) interview techniques. In the first of these studies, Geiselman et al. (1984) obtained 11.00 correct details in response to open-ended questions about characteristics of a person from witnesses in- structed with the cognitive interview, compared with 7.38 details by witnesses in a stan- dard interview condition. Importantly, the cognitive interview did not lead to an increase in incorrect details. Whereas this basic pattern of results has been confirmed in studies with real witnesses (Fisher, Geiselman, & Amador, 1989), other studies have noted an increase in the recall of incorrect details gathered with the cognitive interview. For ex- ample, a study by Finger and Pezdek (1999) found that the cognitive interview increased the recall of both correct and incorrect facial descriptors when compared with a stan- dard interview procedure. Confirming this pattern, a recent meta-analysis of 42 studies by Koehnken, Milne, Memon, and Bull (1999) revealed a large increase in the number of correct details elicited by the cognitive interview and a smaller, yet significant, in- crease in the number of incorrect details elicited. Furthermore, the meta-analysis indi- cated that accuracy rates elicited with the cognitive interview were about the same as accuracy rates achieved with traditional interview methods (84% vs. 82%, respectively). It should be noted that the majority of studies examining the cognitive interview have not focused on obtaining person descriptions per se, so further research in this direction seems worthwhile.
THE RELATIONSHIP BETWEEN PERSON
DESCRIPTIONS AND EYEWITNESS IDENTIFICATION
So far our discussion has focused on the nature and quality of person descriptions. An important related issue involves the relationship between the description and identifica- tion of faces. This in turn leads to two distinct (albeit related) questions. First, what is
- PERSON DESCRIPTIONS 19
of individual features, the veracity of the witnesses’ memory for those features (as re- vealed by the quality of their descriptions) becomes predictive of their recognition per- formance. This finding also potentially offers a key for understanding why face descrip- tion quality bears so little relationship to identification performance with own-race faces—namely, the two tasks may draw on fundamentally different types of knowledge, with the former depending on participants’ memory for distinctive features and the lat- ter depending on their nonverbal knowledge of the face in its entirety (see Farah et al., 1998; Wells & Turtle, 1987). A second exception to the typical absence of a relationship between description quality and face recognition quality comes from studies that have compared the relative ease with which different faces can be described versus recognized. Wells (1985) showed participants multiple faces and then examined their ability to both describe and recog- nize each face. He found that distinctive faces tended to be easier to describe and to rec- ognize than less distinct faces, thereby leading to a modest relationship between recog- nition accuracy and description quality ( r .27) across faces. Although this modest correlation does suggest that certain distinctive faces can be recognized on the basis of individual features, it certainly does not undermine the more common conclusion that typically little relationship between verbal description quality and recognition accuracy can be expected. A final exception to the absence of a relationship between description quality and recognition performance has been observed in studies in which participants were forced to generate rather elaborate descriptions of faces and were later asked to identify these individuals in a lineup identification task (cf. Meissner, Brigham, & Kelley, 2001). In these studies, it appears that the elicitation of elaborate verbal descriptions may lead participants to generate inaccurate details, which then impairs their recognition perfor- mance. Indeed, several studies using such a paradigm (Finger & Pezdek, 1999; Meissner, 2002; Meissner et al., 2001) have found that incorrect details reported in participants’ descriptions are predictive of subsequent identification errors. In short, it seems that despite the clear intuition that witnesses who are better at describing a target should also be better at recognizing it, this relationship has proved to be quite elusive and generally weak. Though the absence of such a relationship may undermine this frequently relied-upon method for assessing the credibility of witnesses, it also provides an important link in our understanding of the nature of person descrip- tions—namely, that person descriptions may draw upon knowledge or cognitive pro- cesses that are very different from those invoked in the identification of a face. More specifically, person descriptions appear to encourage a focus upon verbalizable features of the face that are not always useful for perceptually individuating a given face from among similar distractors. In contrast, recognition of faces has been shown to involve a configural process in which features combine to create a nonverbalizable perceptual set that is stored and later accessed for pattern recognition (Farah et al., 1998). The ex- ceptions to the incompatibility of these processes appear to involve faces that are rec- ognizable based upon a distinctive local feature, or conditions in which retrieval of a face description distorts the veracity of the memory trace and interferes with subsequent identification.
- PERSON DESCRIPTIONS 21
The Influence of Person Descriptions on Identification: Verbal Overshadowing
The fundamental difference between describing a face and recognizing it also contributes to some counterintuitive findings regarding the influence of verbally describing a face on subsequent recognition of that face. Intuitively we might expect that describing a face would be helpful for subsequent memory performance, because it constitutes a form of verbal rehearsal, and verbal rehearsal is well known to enhance memory performance (e.g., Darley & Glass, 1975; Glenberg & Adams, 1978; see Sporer, 1989). There is some evidence that visually rehearsing a face, even after being prompted by a verbal descrip- tion cue, may indeed improve recognition (Sporer, 1988). However, a growing body of re- search suggests that contrary to this intuition, efforts to describe a previously seen face can actually impair subsequent memory performance, at least under some circumstances. In the original documentation of this counterintuitive effect of verbal description on face recognition (termed verbal overshadowing ), Schooler and Engstler-Schooler (1990) showed participants a videotape of a bank robbery. Some participants were instructed to describe the robber in as much detail as possible while others engaged in an unrelated filler activity. Finally, all participants were shown a lineup containing the robber and seven foils. The results revealed that participants who had described the robber were markedly less accurate in recognizing him compared with no-description controls. Fol- low-up experiments by Schooler and Engstler-Schooler were largely consistent with the verbal overshadowing hypothesis that the negative effects of verbalization were due to a mismatch between the visual information or processes associated with the original expe- rience and the verbal information or processes associated with the act of verbal descrip- tion. For example, the negative effects of verbal description generalized to another type of nonverbal stimuli (i.e., colors), but not to more readily verbalized stimuli (i.e., the con- tents of what the robber said). Similarly, whereas verbal rehearsal repeatedly disrupted performance, visualizing the robber’s face had no effect on subsequent identification. Since its original demonstration, the verbal overshadowing phenomenon has been replicated numerous times (Dodson, Johnson, & Schooler, 1997; Fallshore & Schooler, 1995; Ryan & Schooler, 1998; Schooler, Ryan, & Reder, 1996; Sporer, 1989). At the same time, however, it has also failed to replicate on a number of occasions (Lovett, Small, & Engstrom, 1992; Yu & Geiselman, 1993). A meta-analysis of the verbal overshadowing effect was recently conducted by Meissner and Brigham (2001a). Across a sample of 15 studies (29 effect size comparisons; N 2018), Meissner and Brigham observed a small, yet significant, verbal overshadowing effect (Z r .12) demonstrating that par- ticipants who described a target face were 1.27 times more likely to later misidentify the face from a lineup recognition task when compared with participants who did not gen- erate a description prior to identification. Although the verbal overshadowing effect is a reliable phenomenon, it nevertheless appears to be somewhat fragile. Moreover, while research following the original dem- onstration of verbal overshadowing is largely (if not entirely) consistent with the claim that it is associated with discrepancies between the modality of the original visual en- coding, the precise mechanism responsible for the effect remains an issue of some con-
22 MEISSNER, SPORER, SCHOOLER
