An Investigation of the Weapon Focus Effect and the Confidence–Accuracy Relationship for Eyewitness Identification

Journal of Applied Research in Memory and Cognition 6 (2017) 82–92 Contents lists available at ScienceDirect

Journal of Applied Research in Memory and Cognition journal homepage: www.elsevier.com/locate/jarmac

An Investigation of the Weapon Focus Effect and the Confidence–Accuracy Relationship for Eyewitness Identification夽 Curt A. Carlsona,∗ , Jennifer L. Diasa , Dawn R. Weatherfordb , Maria A. Carlsona a

Texas A&M University-Commerce, United States b Arkansas State University, United States

Eyewitness memory can be negatively influenced by the presence of a weapon during a crime. We investigated the potential impact of weapon presence on the confidence–accuracy relationship. Additionally, we tested a concealed weapon condition, as it is common for criminals to verbally threaten a victim with a weapon, despite not showing one during a crime. In support of the weapon focus effect (WFE), correct identifications were lower, and false identifications were higher, for participants who saw the weapon. The concealed weapon did not create a WFE, even though the perpetrator attempted to draw attention to the gun in his pocket, and participants reported that he had a gun. Calibration analyses revealed that weapon presence, whether visible or concealed, did not negatively impact the confidence–accuracy relationship. In fact, participants were best calibrated when the weapon was clearly visible. We discuss implications of these findings for police and the criminal justice system.

General Audience Summary We conducted a large experiment to investigate an important issue concerning eyewitness identification. When a culprit commits a crime with a weapon, it has been shown that eyewitnesses tend to focus on the weapon, thereby harming their memory for the perpetrator. However, it is not understood how confident eyewitnesses are when they make a lineup decision following a crime involving a weapon. It is possible that they could appreciate the fact that they were distracted by the weapon, and adjust their confidence accordingly. This would lead eyewitnesses to be well calibrated, insofar as their confidence could match up well with their identification accuracy. Though confidence is not a perfect indicator of accuracy, it does provide useful information. Prior work has shown that confidence recorded immediately after a lineup decision is moderately strongly correlated with accuracy, at least for fairly pristine encoding situations (e.g., a laboratory setting). We wanted to determine if this relationship is influenced by weapon presence during a mock crime. We investigated eyewitness confidence and accuracy by presenting a mock crime video (purse-snatching) to a large group of undergraduate students as well as a representative sample of participants from across the country (total sample size of 1234). Each participant took part in the experiment online via a computer. Shortly after watching a mock crime in which a culprit either shows a handgun, has no handgun, or conceals the gun in his pocket, participants attempted to identify him from a lineup (which either contained his mugshot or not). Results indicated that confidence can meaningfully be used to distinguish between accurate and inaccurate

Author Note This work was funded by the Office of Research and Sponsored Programs at Texas A&M University-Commerce. 夽 Please note that this paper was handled by the current editorial team of JARMAC.

∗

Correspondence concerning this article should be addressed to Curt A. Carlson, Department of Psychology, Counseling, and Special Education, Texas A&M University-Commerce, Commerce, TX 75429, United States. Tel.: +1 903 468 8723; fax: +1 903 886 5510. Contact: [email protected].

WEAPON CONFIDENCE ACCURACY

83

eyewitnesses, and this relatively strong relationship between confidence and accuracy was present regardless of weapon presence or concealment. In other words, our participant–eyewitnesses who indicated high confidence after choosing from a lineup tended to be accurate, regardless of our weapon manipulations. In addition, the most calibrated participants were actually those who viewed the mock crime with the weapon visible. In conclusion, police should always collect confidence after an eyewitness’s lineup decision, and might not want to devalue identifications after crimes involving a weapon, as long as these identifications are supported by high confidence. Keywords: Eyewitness identification, Weapon focus effect, Weapon concealment, Confidence and accuracy

In terms of eyewitness identification, weapon involvement during a crime is important because eyewitnesses can be susceptible to a weapon focus effect (WFE), whereby their attention is drawn to a weapon rather than other details, such as the perpetrator’s face (e.g., Loftus, Loftus, & Messo, 1987; see meta-analyses by Fawcett, Russell, Peace, & Christie, 2013; and Steblay, 1992). Weapon presence frequently weakens some memories for the crime, such as for perpetrator clothing and other peripheral contextual details. Lineup performance is less consistently impacted, with some studies finding no effect (e.g., Cutler & Penrod, 1988; Cutler, Penrod, O’Rourke, & Martens, 1986) and others showing lower accuracy (e.g., Carlson & Carlson, 2012, 2014; Erickson, Lampinen, & Leding, 2014; O’Rourke, Penrod, Cutler, & Stuve, 1989). The present study focused on the confidence–accuracy relationship, comparing a visible weapon condition with both a weapon absent and a concealed weapon condition. The Federal Bureau of Investigations (2011) reported that, out of 5086 bank robberies, a weapon was threatened 2331 times, but only shown 1242 times. In other words, a perpetrator was almost twice as likely to imply that he had a weapon than to actually show it. This study serves as the first to assess the potential impact of weapon presence and concealment on the eyewitness confidence–accuracy relationship. Before we describe the current state of the literature regarding this relationship, we briefly discuss research on the WFE and previous manipulations of weapon visibility. The Weapon Focus Effect Loftus et al. (1987) conducted the first controlled experiment investigating weapon presence and eyewitness identification (cf. Johnson & Scott, 1976). Participants viewed a series of slides portraying individuals waiting in line at a fast food restaurant. Four critical slides showed either a person handing the cashier a check or pointing a gun at the cashier. After a brief retention interval, participants completed a questionnaire followed by a fair perpetrator-present 12-person lineup. In Experiment 1, memory was marginally worse, and the perpetrator was chosen marginally less, by those in the weapon condition. In Experiment 2, improved power revealed that those in the weapon condition indeed had worse memory than those in the control condition, indicated by both the memory questionnaire and lineup accuracy. Eye-tracking data from Experiment 1 revealed that participants fixated longer on the weapon than the check (see also Biggs, Brockmole, & Witt, 2013). In other words, additional visual

attention on the weapon lowered time spent on other aspects of the crime, including the perpetrator’s face. In response to both types of evidence, Loftus et al. (1987) outlined two potential explanations for weapon focus: arousal and unusualness. The arousal hypothesis states that seeing a weapon increases physiological arousal and stress, which causes tunnel vision on the weapon, akin to Easterbrook’s (1959) cue utilization theory. Peters (1988) found some evidence for this theory by showing that participants were worse at identifying a nurse who administered an injection to them (threat condition), compared to identifying a researcher who provided a questionnaire (but see Maass & Kohnken, 1989). In addition, the WFE appears to be larger under threatening or arousing circumstances (Fawcett et al., 2013). However, though there is evidence for an overall WFE across archival and field studies, the evidence is inconsistent across individual studies (Fawcett et al., 2013), leading several researchers to argue against the idea that lab studies can adequately replicate real world circumstances, including weapon presence (e.g., Cutshall & Yuille, 1989). As an alternative to the arousal hypothesis, the unusualness hypothesis states that a weapon draws attention because it is unexpected within a given context. Pickel (1998, 1999) and others have found a weapon focus effect for objects that are unexpected in a given context, such as a feather duster in a bank (Hope & Wright, 2007), a stick of celery during a business transaction (Mitchell, Livosky, & Mather, 1998), and a raw chicken in a hair salon (Pickel, 1998). Additionally, the weapon focus effect can be eliminated by making a weapon more congruent with surrounding context, such as a gun presented at a shooting range compared to a baseball field (Pickel, 1999). The gun is expected at a shooting range, and because it is no longer unusual in this context, it does not draw more attention than other objects. Manipulations of Weapon Visibility Cutler, Penrod, and colleagues published several studies in the late 1980s investigating a large number of eyewitness identification variables, including weapon visibility (Cutler & Penrod, 1988; Cutler et al., 1986; Cutler, Penrod, & Martens, 1987a; Cutler, Penrod, & Martens, 1987b; O’Rourke et al., 1989; see also Kramer, Buckhout, & Eugenio, 1990). However, neither the WFE nor weapon visibility in particular was the primary focus of any individual study. Each experiment presented the same one or two mock crime videos, both involving a robbery. Evidence for an effect of weapon visibility was mixed. Two studies

WEAPON CONFIDENCE ACCURACY

found that a highly visible weapon led to significantly worse identification accuracy (Cutler et al., 1987a; O’Rourke et al., 1989), two studies found either no effect or a marginal disadvantage for the high visibility condition (Cutler et al., 1986, 1987b), and another study found a numerical, but non-significant, advantage for the high visibility condition (Cutler & Penrod, 1988). Several characteristics of these studies could explain the inconsistencies, which we address in the present study. First, weapon visibility was not clearly operationalized across studies. The high visibility condition is consistently described as the perpetrator outwardly brandishing a handgun during the entirety of the liquor store robbery. However, the low visibility condition description differs slightly, but importantly, across studies. Cutler et al. (1986) stated that “the weapon remained hidden in the robber’s jacket throughout most of the robbery” (p. 117), and Cutler et al. (1987a) also stated that the handgun is hidden under the coat most, but not all of the time. O’Rourke et al. (1989) described both high and low visibility conditions as ending the same way: the robber “threatened the clerk before leaving with the money” (p. 387). In Cutler et al. (1986, 1987a), the implication is that this threat occurs in the low visibility condition with the robber revealing the weapon from under his jacket. However, this same condition is described as involving the weapon hidden under his jacket during the entire video in three of the studies (Cutler & Penrod, 1988; Cutler et al., 1987b; O’Rourke et al., 1989). We attempted to clarify this issue by presenting a new mock crime video with three conditions: (a) no weapon, (b) handgun clearly shown by perpetrator throughout robbery, and (c) handgun completely hidden in perpetrator’s pocket, with his hand in the pocket, and he states that he has a gun. A second characteristic of these weapon visibility studies (e.g., Cutler & Penrod, 1988; Cutler et al., 1986, 1987a, 1987b; O’Rourke et al., 1989) is the use of a fractional factorial design to test several variables with only a limited number of participants per cell. Although this approach has certain advantages (Kenny, 1985), not least of which is a dramatic savings on the number of participants needed per experiment, it produces significant problems (Box, Hunter, & Hunter, 2005). These studies mimic the vast majority of the WFE literature (e.g., Kramer et al., 1990; Loftus et al., 1987) in the use of relatively small sample sizes. To remedy this situation and allow for reasonable power (Aldrich & Nelson, 1984), we utilized a full factorial design with at least 200 participants per cell (also to allow for calibration analyses described below). Moreover, WFE researchers have recently noted a limited number of studies testing the WFE with both perpetrator-present and -absent lineups (Carlson & Carlson, 2012, 2014; Erickson et al., 2014). We included both lineup types in order to present a more complete picture of eyewitness identification, and also to allow for calibration analysis. Taken as a whole, it is not clear what prediction comes from these weapon visibility studies regarding the weapon concealment manipulation in the present study. However, it is reasonable to hypothesize that a concealed weapon condition might not produce a WFE, which would not be surprising if participants are not aware of the weapon. It is unclear from the description of the low visibility condition in the above studies whether or not the perpetrator intentionally drew attention to the weapon in his pocket,

84

and if so, whether this occurred at the very end of the crime, or throughout the crime. For the present study, we decided that the best comparison is between a visible weapon and a concealed weapon condition in which attention is intentionally drawn to the weapon. This latter situation reflects real world crimes involving concealed weapons, in that criminals would likely call some attention to the weapon in order to gain compliance from a victim (e.g., for a cashier to hand over money), but not explicitly show the weapon, so as not to draw attention from bystanders and security personnel. In addition, Fawcett et al. (2013), after their meta-analysis of the WFE literature, made note of a lack of research comparing a concealed weapon with both a visible weapon and a weapon absent condition. We sought to fill this gap in the literature, and also to extend the WFE literature by focusing on the confidence–accuracy relationship. The Confidence–Accuracy Relationship There is a long history of research on eyewitness confidence and accuracy (CA), a review of which is beyond the scope of this paper. However, we will briefly describe how some of the conclusions have changed. In an early review, Wells and Murray (1984) found a small point–biserial correlation coefficient of .07, concluding that the CA relationship is weak. Other researchers reached the same conclusion in following years (e.g., Kassin, Ellsworth, & Smith, 1989; Smith, Kassin, & Ellsworth, 1989). Later it was shown that the CA relationship is much stronger for choosers (i.e., those who identify an individual from a lineup); the point–biserial correlation jumped to .41 (see meta-analysis by Sporer, Penrod, Read, & Cutler, 1995). Shortly after this meta-analysis, Juslin, Olsson, and Winman (1996) showed that calibration statistics reveal much more about the CA relationship than the point–biserial correlation coefficient. At a minimum, researchers can depict calibration curves representing changes in proportion correct across levels of confidence. In this way, it can easily be determined whether, for example, eyewitnesses who express 90–100% confidence are highly accurate with their lineup decisions. Since the pivotal study by Juslin et al. (1996), several studies have focused on CA calibration (e.g., Brewer, Keast, & Rishworth, 2002; Brewer & Wells, 2006; Horry, Palmer, & Brewer, 2012; Keast, Brewer, & Wells, 2007; Palmer, Brewer, Weber, & Nagesh, 2013; Sauer, Brewer, Zweck, & Weber, 2010; Sauerland & Sporer, 2009; Sauerland, Sagna, & Sporer, 2012; Weber & Brewer, 2004; Weber & Brewer, 2006; see Wixted, Mickes, Clark, Gronlund, & Roediger, 2015, for a review). It is now established that the CA relationship is relatively strong, as long as it is based on confidence assessed immediately after the identification decision, and collected under favorable conditions (e.g., attentive participants after a fairly short retention interval). Predictions Though some researchers have analyzed eyewitness confidence after a weapon manipulation (e.g., Hope & Wright, 2007), there has been no research investigating the impact of weapon presence on the CA relationship. It would be useful for police to understand the potential impact of a perpetrator’s weapon on

WEAPON CONFIDENCE ACCURACY

85

eyewitness confidence, and it might intuitively be expected that the CA relationship would be weaker if there was a weapon. In contrast, we predict that a visible weapon would not harm the CA relationship, as participants should realize that their memory for the perpetrator’s face could be negatively influenced, therefore adjusting their confidence accordingly. In other words, we expected that participants would realize that they are distracted by the weapon, and would metacognitively appreciate the deleterious impact of this distraction on their memory for the perpetrator. This metacognitive awareness should lead to an adjustment of their confidence in their lineup decision. A related idea is that eyewitnesses who view a perpetrator for a very short period of time should be less confident in their ability to identify him (Memon, Hope, & Bull, 2003). Read (1995) described a related scenario to explain why eyewitnesses who view a perpetrator for a relatively long period of time can counterintuitively produce elevated false alarm rates (e.g., Shapiro & Penrod, 1986). The argument is that eyewitnesses can be fooled into thinking that they remember the perpetrator much better than they actually do, and are much more willing to choose from a lineup as a result, thereby increasing false identifications. We argue that metacognition can also have a beneficial effect, insofar as eyewitnesses could realize that their memory for the perpetrator would be worse when a weapon is present (for more information about eyewitness metacognition, see Hollins & Weber, 2016). A similar pattern could arise even when the weapon is concealed, as long as participants notice that the perpetrator apparently has a gun in his pocket. As described below, we designed our concealment condition to draw attention to the concealed weapon, and included a manipulation check to determine whether or not participants believed that the perpetrator had a weapon in his pocket. We make no claims about attentional capture or dwell, as we did not collect eye-tracker data to test such hypotheses. Rather, we simply tested the possibility that a concealed weapon could produce downstream cognitive consequences similar to a visible weapon, in terms of eyewitness identification (i.e., the WFE) and the CA relationship. In contrast, the research on weapon visibility indicates that our concealment condition might not produce a WFE. If this is the case, than the CA relationship should be similar between the concealed and no weapon conditions. It is certainly possible that participants must have an explicit weapon on which to focus in order to produce a weapon focus effect.

obtained ethical approval from each university’s institutional review board prior to conducting the experiment. There were no significant differences between the university samples and nationwide sample, so we collapsed across all samples for analyses below (total N = 1234).

Method

Following informed consent, participants were instructed to pay close attention to the mock crime video. They did not know that a crime would be depicted, or that their memory would later be tested. After watching the video, participants spent 5–10 min working through anagrams of U.S. States. Following this task, they were instructed, “On the next screen is a lineup that may or may not contain the perpetrator from the video. Your goal is to identify the perpetrator if he is present, otherwise press ‘n’ to indicate ‘none of the above.”’ They were then presented with either the PP or PA simultaneous lineup. For convenience, we simply counterbalanced perpetrator position in either the top middle or bottom middle of the 2 × 3 photospread. Participants

Participants Undergraduate students (N = 655; 70% female; M age = 21.0, SD = 5.8) from either Texas A&M University-Commerce or Arkansas State University participated in this study, in partial fulfillment of course credit requirements. A nationwide sample from SurveyMonkey provided an additional 579 participants (49% female; M age = 31.9, SD = 12.3). This study was listed among several online surveys in which SurveyMonkey members could participate for credits that they could later spend on merchandise. There was no explicit advertising for our study. We

Materials We recorded three versions of a mock crime with a digital camera attached to a tripod for a third-person perspective. All videos begin with a young Caucasian woman talking on her phone and walking to the door of her apartment. As she begins to look for her keys in her purse a young Caucasian male comes up from behind her and speaks the following lines in each condition: “Hey, don’t make a sound. Give me your money.” In both weapon shown and weapon concealed conditions, between these two sentences, he also states, “I’ve got a gun.” The perpetrator then grabs her purse and turns and runs away. The same scenario is used for all conditions, and each lasts approximately 13 s. The perpetrator’s face is clearly visible from approximately 1.5 m away during the video (see Figure 1 for static images). In one version, the perpetrator has no weapon. In another version, the weapon is fully visible in his hand. In the third condition he refers to the weapon held in his hand in his external jacket pocket visible to the camera (though no part of the weapon is visible). The culprit stood for a mugshot a few days later, wearing different clothing. This mugshot was presented to five research assistants to establish the following modal description: Caucasian male, early 20 s, short brown hair, approximately 1.8 m tall. We created several lineups with mugshots obtained based on this description from the Kentucky Department of Corrections database (http://apps.corrections.ky.gov/ KOOL/AdvancedSearch). A group of 30 undergraduate students, none of whom later took part in the experiment, chose a member from each lineup who they felt best matched the modal description. Using these blind lineup selections, we chose the perpetrator-present (PP) lineup and the perpetrator-absent (PA) lineup with the highest effective size based on Tredoux’s E (Tredoux, 1998). This measure ranges from 0 to the nominal lineup size, with higher numbers indicating a fairer lineup. The PA lineup yielded a value of 4.25 (95% CIs: 3.48, 5.43), and the PP lineup yielded a value of 5.36 (95% CIs: 4.38, 6.89). Procedure

WEAPON CONFIDENCE ACCURACY

86

Figure 1. The top two images are from the weapon visible condition, the bottom-left image shows the concealed weapon, and the bottom-right image is from the no weapon condition. The camera was the same distance from the perpetrator for each condition, and our actors performed the same scripted movements. Table 1 Identification and Rejection Rates Across Conditions Condition

No weapon Weapon concealed Weapon shown

Perpetrator-present lineup

Perpetrator-absent lineup

Correct ID rate

Foil ID rate

Rejection rate

Foil ID rate

Rejection rate

.50 (104/206) .46 (97/212) .25 (51/206)

.29 (60/206) .33 (69/212) .43 (88/206)

.20 (42/206) .22 (46/212) .33 (67/206)

.74 (153/203) .72 (149/206) .84 (169/201)

.25 (50/203) .28 (57/206) .16 (32/201)

Notes: ID, identification; raw proportions based on number of participants per condition in parentheses. Not all proportions add to 1.0 due to rounding.

pressed a number 1–6 to choose one of the six lineup members (a numerical label was beneath each photo) or “n” to indicate none of the above. Immediately following their decision, participants rated their confidence on a 0–100% Likert scale with 10-point intervals (e.g., 0, 10, 20, . . ., 100%). Design and Analysis We utilized a 3 (video: weapon shown, weapon concealed, or no weapon) × 2 (perpetrator-present [PP] or -absent [PA] lineup) between-subjects factorial design. Two measures of accuracy result from this design: correct identifications of the perpetrator from the PP lineup, and foil identifications from the PA lineup, each of which we analyzed with logistic regression. The confidence judgments were used to conduct calibration analyses (described in detail below).1

1 When eyewitness decisions and confidence from PP and PA lineups are available, these can also be used for receiver operating characteristic (ROC)

Results Logistic Regression See Table 1 for the frequencies and proportions across all conditions for both PP and PA lineup decisions. Using SPSS we conducted separate logistic regression analyses (see Carlson & Carlson, 2014). The first involved only the perpetrator-present (PP) lineups, comparing the number of correct identifications (IDs) of the perpetrator (coded as 1) versus all other decisions (coded as 0), across the three weapon conditions. We then

analysis, which reveals which conditions create the best ability of eyewitnesses to discriminate between innocent and guilty suspects (e.g., Gronlund, Wixted, & Mickes, 2014; Mickes, Flowe, & Wixted, 2012; National Research Council, 2014; Wixted & Mickes, 2012, 2015). We conducted ROC analysis of the present data, and these results are presented in an appendix along with a full breakdown of participant decisions across levels of confidence. We did not include these results in the main body of the manuscript because our focus is the confidence–accuracy relationship rather than discriminability.

WEAPON CONFIDENCE ACCURACY

analyzed the number of foil IDs for the PP lineup (coded as 1), versus all other decisions (coded as 0). The last analysis of the PP data was of rejections (coded as 0), compared to all IDs (coded as 1). As for the perpetrator-absent (PA) data, we compared the number of foil IDs (coded as 1) versus lineup rejections (coded as 0). We next describe the PP data, followed by the PA data. All p-values are 2-tailed. The full model PP logistic regression was significant for correct IDs, Wald(2) = 30.98, p < .001. The weapon shown condition yielded significantly fewer correct IDs compared to both the weapon concealed (Wald(1) = 19.67, p < .001) and the no weapon condition (Wald(1) = 28.13, p < .001). There was no significant difference in the number of correct IDs between weapon concealed and no weapon conditions, Wald(1) = 0.94, p = .33. For foil IDs, the full model was also significant, Wald(2) = 9.01, p = .01. There was no significant difference between no weapon and concealed conditions, Wald(1) = 0.57, p = .45, but there were significantly more foil IDs in the weapon shown condition compared to both no weapon condition (Wald(1) = 8.19, p = .004) and concealed condition (Wald(1) = 4.59, p = .032). Turning to PP lineup rejections, the full model again was significant, Wald(2) = 9.69, p = .008. There was no significant difference between no weapon and concealed conditions, Wald(1) = 0.11, p = .74, but there were significantly more lineup rejections in the weapon shown condition compared to both no weapon condition (Wald(1) = 7.69, p = .006) and concealed condition (Wald(1) = 6.14, p = .013). The full model PA logistic regression also was significant, Wald(2) = 8.40, p = .015. The weapon shown condition produced more foil IDs than both the weapon concealed (Wald(1) = 8.05, p = .005) and the no weapon condition (Wald(1) = 4.68, p = .031). There was no significant difference in foil IDs between no weapon and weapon concealed conditions, Wald(1) = 0.49, p = .49. These results indicate that when a culprit clearly shows a weapon during a crime, this reduces the ability of eyewitnesses to later identify the culprit, and also increases the likelihood that eyewitnesses will identify a foil from a perpetrator-absent lineup. However, concealing the weapon apparently does not harm eyewitness identification from either PP or PA lineups. Receiver operating characteristic analysis confirmed that the ability of our participants to discriminate between the perpetrator and innocent suspects was reduced significantly in the visible weapon condition, compared to both the no weapon and concealed conditions (see Appendix). There was no difference in objective discriminability between no weapon and concealed conditions. Calibration Our primary aim was to investigate the potential influence of weapon presence and concealment on the confidence–accuracy (CA) relationship. We predicted that, even though a visible weapon reduces identification accuracy, eyewitnesses might be metacognitively aware that the weapon could be harming their memory of the perpetrator, thereby adjusting their confidence

87

Figure 2. Calibration curves from all conditions. Proportion correct represents the probability that a suspect is guilty, given that he was chosen from a lineup, and it is calculated as: number of correct identifications from perpetrator-present lineups divided by number of correct identifications from perpetrator-present lineups plus estimated number of innocent suspect identifications from perpetratorabsent lineups. The number of innocent suspect identifications is estimated by dividing the total number of identifications from perpetrator-absent lineups by the nominal size of the lineup, six. Bars represent standard error. The thin diagonal line represents perfect calibration.

accordingly. This adjustment should keep confidence and accuracy reasonably calibrated, which would be good news to police, as they could still trust high-confident identifications, even after a crime involving a weapon. A useful first step when investigating calibration is to produce calibration curves (see Figure 2). As mentioned before, the CA relationship is much stronger for choosers than for nonchoosers (e.g., Sporer, 1992; Sporer et al., 1995; see also Brewer & Wells, 2006), therefore the accuracy measure depicted on the y-axis is proportion correct based only on identifications (rather than correct rejections of PA lineups). Here proportion correct is defined as: correct IDs/(correct IDs + false IDs), with correct IDs being perpetrator IDs from PP lineups, and false IDs being all foil IDs from PA lineups divided by the nominal lineup size (six in this case; see Mickes, 2015). It is common to collapse confidence into five bins on the x-axis (e.g., Palmer et al., 2013). The thin straight diagonal line represents perfect calibration, such that each level of confidence equals the level of accuracy. When a curve strays above the diagonal, this indicates under-confidence, and when a curve extends below, eyewitnesses are over-confident. A cursory evaluation of the curves reveals that participants were fairly well calibrated across all three conditions. When the weapon was concealed or absent, participants were generally under-confident, but this tendency was reduced when the weapon was shown. In fact, it appears as though the weapon shown condition was best calibrated, as this curve lies closest to the line of perfect calibration. In order to address this possibility, and also to more concretely ascertain the level of over- versus under-confidence for each condition, we calculated calibration statistics. There are three commonly used calibration statistics. Calibration (C) represents how close each curve is to the line of perfect calibration across the entire confidence range. It ranges from 0 (perfect calibration) to 1. A related statistic is over/under

WEAPON CONFIDENCE ACCURACY Table 2 Calibration Statistics with Inferential 95% Confidence Intervals Condition

C

O/U

Resolution

No weapon ICI Concealed ICI Shown ICI

.026 .019, .033 .029 .021, .037 .007 .003, .011

−.275 −.263, −.287 −.260 −.246, −.274 −.165 −.137, −.193

.270 .226, .314 .241 .195, .287 .293 .225, .361

Notes: ICI, inferential confidence interval (α = .05); C, calibration, with lower numbers representing better calibration; O/U, over/under confidence, with negative numbers representing under-confidence, and positive numbers indicating over-confidence. For resolution, higher numbers indicate a better ability of confidence to discriminate between correct and incorrect identifications.

confidence (O/U), which simply shows how much each curve strays above versus below the perfect calibration line, again across all confidence levels. It ranges from −1 (very underconfident) to 1 (very over-confident). The third statistic, resolution, can be more difficult to assess from a visual inspection of calibration curves. Resolution is the ability of confidence ratings in a given condition to postdict accuracy. In other words, it is arguably the most valuable calibration statistic, as it informs us how well confidence discriminates between accurate and inaccurate identifications. Resolution can be calculated in various ways; we chose the Normalized Discrimination Index (Yaniv, Yates, & Smith, 1991). See Brewer and Wells (2006) for a detailed description of each of these statistics. Following Palmer et al. (2013), we used a modified jackknife procedure (Mosteller & Tukey, 1968) to generate standard errors for each statistic for each condition. We then converted each standard error to inferential confidence intervals (Palmer, Brewer, & Weber, 2010; Tryon, 2001) based on an alpha level of .05. If there is no overlap in confidence intervals, this indicates a significant difference. The three calibration statistics with inferential CIs are depicted in Table 2. Eyewitness confidence and accuracy were best calibrated in the weapon shown condition, with no significant difference between the no weapon and concealed conditions. All conditions produced under-confidence, an issue we will address in “Discussion” section. However, participants in the weapon shown condition were less under-confident, which corresponds with the fact that they were better calibrated. Finally, our estimate of resolution can be interpreted as an effect size, such that a small effect is around .01, a medium effect around .06, and a large effect around .14 (Brewer & Wells, 2006). As evidenced in Table 2, confidence judgments made by participants in all conditions had a large effect on resolution (i.e., discriminating between accurate and inaccurate suspect choices), as all resolution estimates are well above .14. Interestingly, weapon presence did not impact this ability.

88

She recommended plotting confidence–accuracy characteristic (CAC) curves (Figure 3). From a methodological standpoint, CAC plots are useful when calibration statistics cannot be computed, due to insufficient sample sizes or a confidence scale that does not map directly onto accuracy (e.g., Likert scale of 1–7). CAC space is virtually identical to calibration space, especially when proportion correct (the y-axis in both cases) is calculated the same way: # correct IDs/(# correct IDs + # false IDs), which is the probability that the suspect is guilty, given a suspect ID. Like our calibration analysis, we continue to use this calculation, thereby focusing on suspect IDs by including only perpetrator IDs (i.e., correct IDs) from the PP lineup, and dividing false IDs from the PA lineup by the nominal lineup size of six to arrive at an estimate of innocent suspect IDs (which are more comparable to correct IDs than including all foil IDs). It is particularly important for triers of fact that CAC analysis focuses on suspect IDs rather than foil IDs (Mickes, 2015). One difference between calibration and CAC analysis is that, in CAC space, the diagonal line of perfect calibration is not easy to depict, as the confidence scale does not need to match up directly with accuracy (i.e., when they are not both on 0–100% scales). Essentially CAC curves simplify evaluation of the CA relationship for triers of fact, as they can simply depict low versus medium versus high confidence, as done in Figure 3. Rather than drawing attention to levels of over- versus under-confidence across the entire confidence range, as calibration curves allow, and which would be important for theoretical evaluations of the CA relationship, these CAC curves focus on the most relevant range of the confidence scale, namely the eyewitnesses who exhibit the highest confidence. These curves address the question: can identifications made by highly confident eyewitnesses (those most likely to make it to trial) be trusted? In other words, are these identifications highly accurate? Figure 3 shows that they are. This pattern of highly confident eyewitnesses choosing accurately from a lineup, regardless of encoding conditions, replicates data by Palmer et al. (2013). They compared 5 s encoding with 90 s encoding, and found that, though discriminability

Confidence–Accuracy Characteristic Curves Mickes (2015) argued that triers of fact in the criminal justice system (e.g., judges and juries) would be particularly interested in a slightly different portrayal of the CA relationship.

Figure 3. Confidence–accuracy characteristic (CAC) curves with standard error bars.

WEAPON CONFIDENCE ACCURACY

was higher for the 90 s condition, participants in both conditions were fairly well calibrated. In other words, high-confidence identifications tended to be accurate, whether memory was weak (5 s encoding) or strong (90 s encoding). Discussion We had two goals with the present study. First, we wanted to extend the weapon focus effect (WFE) literature by comparing a concealed weapon condition with a visible weapon and no weapon condition. If a perpetrator threatens a victim with a supposed weapon in his pocket, does this produce a WFE? Second, we sought to determine the impact of the WFE on the confidence–accuracy (CA) relationship. Eyewitness identification accuracy can be influenced by weapon presence, but can eyewitnesses adjust their confidence after seeing a weapon, in order to remain reasonably calibrated? How does weapon concealment influence the CA relationship? In terms of eyewitness accuracy, results were straightforward. We replicated the WFE by showing that a visible weapon reduced the ability of eyewitnesses to discriminate between a guilty suspect and innocent suspects (Carlson & Carlson, 2014; see ROC analysis in Appendix). Seeing the handgun during the mock crime made participants less able to identify the perpetrator when he was in the lineup. Rather than picking him, participants were more likely to either chose a foil or reject the lineup (see Table 1). Seeing the weapon also led participants to choose a foil from the perpetrator-absent lineup. Interestingly, a concealed gun did not influence either correct or false identifications, even though we were careful to have the perpetrator attempt to draw attention to the gun in his pocket. A manipulation check at the end of the experiment confirmed that the vast majority of participants (94%) in the concealed condition reported that the perpetrator had a gun. Regardless, the concealed condition was no different from the no weapon condition, and both yielded more correct identifications, and fewer false identifications, compared to the visible weapon condition. These findings are broadly in agreement with those by Cutler et al. (1987a) and O’Rourke et al. (1989), who found that a highly visible weapon negatively impacted eyewitness identification compared to a low visibility condition. The similarity between the no weapon and concealed weapon conditions could be viewed as puzzling based on the unusualness hypothesis (e.g., Pickel, 1998), which predicts that a schema-inconsistent object should draw attention away from a perpetrator’s face. Though the gun was not visible in the concealment condition, the perpetrator did threaten the victim by stating that he had a gun, and it was obvious that he was holding something in his pocket, which was clearly visible from participants’ point of view. Of course, the pocket itself is not unusual, nor is having one’s hand in one’s pocket, so our results indicate that apparently a weapon or unusual object must be visible in order to draw attention and create an object/weapon focus effect. We do not view our results as particularly powerful evidence against the unusualness hypothesis, however. Our goal was not to provide a strong test of this hypothesis, but our findings could

89

be interpreted as a potential boundary condition for the WFE. We recommend additional research on weapon concealment in order to determine whether this result will replicate across different crime scenarios. Also, different levels of concealment should be investigated. For example, would a visible gun handle create a WFE, or even just a hint of metal that could be interpreted as a weapon (Fawcett et al., 2013)? Ideally, eye-tracker data would be collected to more concretely address this real-world issue, as well as the more theoretical concern about visual attention. More importantly, we investigated the potential impact of weapon presence on the CA relationship. Should police be skeptical of an eyewitness’s identification if they know that the crime involved a weapon? Or should they trust the identification, especially if the eyewitness assigns a high degree of confidence immediately afterward? Our calibration analyses revealed that, not only did the visible weapon not harm the CA relationship, it actually improved it. Participants were better calibrated, and less under-confident, when the weapon was visible rather than concealed or absent. All conditions yielded high resolution, indicating that confidence discriminated well between accurate and inaccurate lineup decisions, regardless of weapon presence. Why were participants better calibrated after seeing the weapon? This can be broken-down into two questions: (a) what made participants in all conditions under-confident, and (b) why did the visible weapon condition reduce this under-confidence? We propose that overall under-confidence was driven by limited exposure time to the perpetrator during the mock crime. The relatively brief exposure to the perpetrator during the video could have made participants more suspicious of their memory than they should have been (Memon et al., 2003). They might have suspected that their memory was so poor that they performed poorly during the lineup. However, though the correct identification rates were not particularly high (see Table 1), they were well above chance. Participants were better at identifying the culprit than they thought they were, possibly due to the fact that, although the perpetrator’s face was not visible for long, it was clearly visible from a short distance. Why then did the visible weapon significantly reduce under-confidence and increase calibration (see Table 2)? These effects could have been driven by either an increase in confidence for the visible weapon condition, or reduced proportion correct. As evidenced by the logistic regression results, the visible weapon reduced correct identifications and increased false identifications, thereby reducing proportion correct. In essence, this served to reduce accuracy to better match with the low level of confidence. Conclusions and Implications Though the identification accuracy results are fairly clear, their interpretation by those in the criminal justice system might not be, particularly in light of the confidence–accuracy results. Relatively high confidence after choosing from a lineup tended to indicate high accuracy in that decision (Figures 2 and 3),

WEAPON CONFIDENCE ACCURACY

regardless of a weapon being shown, concealed, or absent; lower confidence indicated lower accuracy across conditions. A similar pattern can be found from studies manipulating other estimator variables, such as encoding time (e.g., Palmer et al., 2013). What should be done with eyewitnesses of a crime involving a weapon? Simply put, confidence must be assessed immediately after their lineup decision. Based on the CA results from our experiment, we tentatively argue that police could potentially trust an eyewitness who chooses from a lineup and then immediately supports this decision with high confidence. Unlike confidence expressed well after a lineup decision, which can be influenced by a variety of factors, including feedback from a case detective (e.g., Wells & Bradfield, 1998), confidence assessed immediately after a lineup correlates moderately strongly with accuracy (see review by Wixted et al., 2015). Therefore, as long as confidence is assessed immediately after a lineup decision, and given that our results are replicated outside of the lab, the criminal justice system could assign more weight to testimony from eyewitnesses who indicated high confidence, regardless of certain characteristics of the crime, such as weapon presence. Conflict of Interest Statement The authors declare that they have no conflict of interest. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jarmac. 2016.04.001. References Aldrich, J. H., & Nelson, F. D. (1984). Linear probability, logit and probit models. Beverly Hills, CA: Sage. Biggs, A. T., Brockmole, J. R., & Witt, J. K. (2013). Armed and attentive: Holding a weapon can bias attentional priorities in scene viewing. Attention, Perception and Psychophysics, 75, 1715–1724. http://dx.doi.org/10.3758/s13414-013-0538-6 Box, G. E. P., Hunter, J. S., & Hunter, W. G. (2005). Statistics for experimenters: Design, innovation and discovery (2nd ed.). Hoboken, NJ: Wiley-Interscience. Brewer, N., Keast, A., & Rishworth, A. (2002). The confidence–accuracy relationship in eyewitness identification: The effects of reflection and disconfirmation on correlation and calibration. Journal of Experimental Psychology: Applied, 8, 44–56. http://dx.doi.org/10.1037/1076-898X.8.1.44 Brewer, N., & Wells, G. L. (2006). The confidence–accuracy relationship in eyewitness identification: Effects of lineup instructions, foil similarity, and target-absent base rates. Journal of Experimental Psychology: Applied, 12, 11–30. http://dx.doi.org/10.1037/1076-898X.12.1.11 Carlson, C. A., & Carlson, M. A. (2012). A distinctiveness-driven reversal of the weapon-focus effect. Applied Psychology in Criminal Justice, 8, 35–53. Carlson, C. A., & Carlson, M. A. (2014). An evaluation of lineup presentation, weapon presence, and distinctive feature using ROC analysis. Journal of Applied Research in Memory and Cognition, 3, 45–53. http://dx.doi.org/10.1016/j.jarmac.2014.03.004

90

Cutler, B. L., & Penrod, S. D. (1988). Improving the reliability of eyewitness identification: Lineup construction and presentation. Journal of Applied Psychology, 73, 281–290. http://dx.doi.org/10.1037/0021-9010.73.2.281 Cutler, B. L., Penrod, S. D., & Martens, T. K. (1987a). The reliability of eyewitness identification: The role of system and estimator variables. Law and Human Behavior, 11, 233–258. http://dx.doi.org/10.1007/BF01044644 Cutler, B. L., Penrod, S. D., & Martens, T. K. (1987b). Improving the reliability of eyewitness identification: Putting context into context. Journal of Applied Psychology, 72, 629–637. http://dx.doi.org/10.1037/0021-9010.72.4.629 Cutler, B. L., Penrod, S. D., O’Rourke, T. E., & Martens, T. K. (1986). Unconfounding the effects of contextual cues on eyewitness identification accuracy. Social Behaviour, 1, 113–134. Cutshall, J. L., & Yuille, J. C. (1989). Field studies of eyewitness memory of actual crimes. In D. C. Raskin (Ed.), Psychological methods in criminal investigation and evidence (pp. 97–124). New York, NY: Springer. Easterbrook, J. A. (1959). The effect of emotion on the utilization and organization of behavior. Psychological Review, 66, 183–201. http://dx.doi.org/10.1037/h0047707 Erickson, W. B., Lampinen, J. M., & Leding, J. K. (2014). The weapon focus effect in target-present and target-absent line-ups: The roles of threat, novelty and timing. Applied Cognitive Psychology, 28, 349–359. http://dx.doi.org/10.1002/acp.3005 Fawcett, J. M., Russell, E. J., Peace, K. A., & Christie, J. (2013). Of guns and geese: A meta-analytic review of the ‘weapon focus’ literature. Psychology, Crime & Law, 19, 35–66. Federal Bureau of Investigations. (2011). Violations of the Federal Bank Robbery and Incidental Crimes Statute. Bank Crime Statistics (Title 18, United States Code, Section 2113). Retrieved from. http://www.fbi.gov/stats-services/publications/bank-crimestatistics-2011/bank-crime-statistics-2011 Gronlund, S. D., Wixted, J. T., & Mickes, L. (2014). Evaluating eyewitness identification procedures using ROC analysis. Current Directions in Psychological Science, 23, 3–10. http://dx.doi.org/10.1177/0963721413498891 Hollins, T. J., & Weber, N. (2016). Monitoring and regulation of accuracy in eyewitness memory: Time to get some control. In J. Dunlosky, & S. Tauber (Eds.), The Oxford handbook of metamemory. Oxford University Press. Hope, L., & Wright, D. (2007). Beyond unusual? Examining the role of attention in the weapon focus effect. Applied Cognitive Psychology, 21, 951–961. Horry, R., Palmer, M. A., & Brewer, N. (2012). Backloading in the sequential lineup prevents within-lineup criterion shifts that undermine eyewitness identification performance. Journal of Experimental Psychology: Applied, 18, 346–360. http://dx.doi.org/10.1037/a0029779 Johnson, C., & Scott, B. (1976). Eyewitness testimony and suspect identification as a function of arousal, sex of witness and scheduling of interrogation. Washington, DC: Paper presented at the American Psychological Association annual meeting. Juslin, P., Olsson, N., & Winman, A. (1996). Calibration and diagnosticity of confidence in eyewitness identification: Comments on what can be inferred from the low confidence accuracy correlation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 1304–1316. Kassin, S. M., Ellsworth, P. C., & Smith, V. L. (1989). The “general acceptance” of psychological research on eyewitness

WEAPON CONFIDENCE ACCURACY

testimony. American Psychologist, 44, 1089–1098. http://dx.doi.org/10.1037/0003-066X.44.8.1089 Keast, A., Brewer, N., & Wells, G. L. (2007). Children’s metacognitive judgments in an eyewitness identification task. Journal of Experimental Child Psychology, 97, 286–314. http://dx.doi.org/10.1016/j.jecp.2007.01.007 Kenny, D. (1985). Quantitative methods for social psychology. In G. Lindzey, & E. Aronson (Eds.), The handbook for social psychology (vol. 1) (pp. 487–508). New York: Random House. Kramer, T. H., Buckhout, R., & Eugenio, P. (1990). Weapon focus, arousal, and eyewitness memory. Law and Human Behavior, 14, 167–184. http://dx.doi.org/10.1007/BF01062971 Loftus, E. F., Loftus, G. R., & Messo, J. (1987). Some facts about “weapon focus”. Law and Human Behavior, 11, 55–62. http://dx.doi.org/10.1007/BF01044839 Maass, A., & Kohnken, G. (1989). Eyewitness identification: Simulating the ‘weapon effect’. Law and Human Behaviour, 13, 397–408. http://dx.doi.org/10.1007/BF01056411 Memon, A., Hope, L., & Bull, R. (2003). Exposure duration: Effects on eyewitness accuracy and confidence. British Journal of Psychology, 94, 339–354. Mickes, L. (2015). Receiver operating characteristic analysis and confidence–accuracy characteristic analysis in investigations of system variables and estimator variables that affect eyewitness memory. Journal of Applied Research in Memory and Cognition, 4, 93–102. http://dx.doi.org/10.1016/j.jarmac.2015.01.003 Mickes, L., Flowe, H. D., & Wixted, J. T. (2012). Receiver operating characteristic analysis of eyewitness memory: Comparing the diagnostic accuracy of simultaneous versus sequential lineups. Journal of Experimental Psychology: Applied, 4, 361–376. http://dx.doi.org/10.1037/a0030609 Mitchell, K. J., Livosky, M., & Mather, M. (1998). The weapon focus effect revisited: The role of novelty. Legal and Criminological Psychology, 3, 287–303. http://dx.doi.org/10.1111/ j.2044-8333.1998.tb00367.x Mosteller, F., & Tukey, J. W. (1968). Data analysis including statistics. In G. Lindzey, & E. Aronson (Eds.), The handbook of social psychology (pp. 80–203). Reading, MA: Addison-Wesley. National Research Council. (2014). Identifying the culprit: Assessing eyewitness identification. Washington, DC: The National Academies Press. O’Rourke, T. E., Penrod, S. D., Cutler, B. L., & Stuve, T. E. (1989). The external validity of eyewitness identification research: Generalizing across subject populations. Law and Human Behavior, 13, 385–395. http://dx.doi.org/10.1007/BF01056410 Palmer, M. A., Brewer, N., & Weber, N. (2010). Postidentification feedback affects subsequent eyewitness identification performance. Journal of Experimental Psychology: Applied, 16, 387–398. http://dx.doi.org/10.1037/a0021034 Palmer, M. A., Brewer, N., Weber, N., & Nagesh, A. (2013). The confidence–accuracy relationship for eyewitness identification decisions: Effects of exposure duration, retention interval, and divided attention. Journal of Experimental Psychology: Applied, 19, 55–71. http://dx.doi.org/10.1037/a0031602 Peters, D. P. (1988). Eyewitness memory and arousal in a natural setting. In M. Gruneberg, P. Morris, & R. Skyes (Eds.), Memory in everyday life (vol. 1) Practical aspects of memory: Current research and issues (pp. 89–94). Chichester: Wiley. Pickel, K. L. (1998). Unusualness and threat as possible causes of “weapon focus”. Memory, 6(3), 277–295. http://dx.doi.org/10.1080/ 096582198388292

91

Pickel, K. L. (1999). The influence of context on the “weapon focus” effect. Law and Human Behavior, 23(3), 299–311. http://dx.doi.org/10.1023/A:1022356431375 Read, J. D. (1995). The availability heuristic in person identification – The sometimes misleading consequences of enhanced contextual information. Applied Cognitive Psychology, 9, 91–121. http://dx.doi.org/10.1002/acp.2350090202 Sauer, J. D., Brewer, N., Zweck, T., & Weber, N. (2010). The effect of retention interval on the confidence–accuracy relationship for eyewitness identification. Law and Human Behavior, 34, 337–347. http://dx.doi.org/10.1007/s10979-009-9192-x Sauerland, M., Sagna, A., & Sporer, S. L. (2012). Assessing nonchoosers’ eyewitness identification accuracy from photographic showups by using confidence and response times. Law and Human Behavior, 36, 394–403. Sauerland, M., & Sporer, S. L. (2009). Fast and confident: Postdicting eyewitness identification accuracy in a field study. Journal of Experimental Psychology: Applied, 15, 46–62. http://dx.doi.org/10.1037/a0014560 Shapiro, P., & Penrod, S. (1986). Meta-analysis of facial identification studies. Psychological Bulletin, 100, 139–156. http://dx.doi.org/10.1037/0033-2909.100.2.139 Smith, V. L., Kassin, S. L., & Ellsworth, P. C. (1989). Eyewitness accuracy and confidence: Within- versus between-subjects correlations. Journal of Applied Psychology, 74, 356–359. http://dx.doi.org/10.1037/0021-9010.74.2.356 Sporer, S. L. (1992). Post-dicting eyewitness accuracy: Confidence, decision times and person descriptions of choosers and non-choosers. European Journal of Social Psychology, 22, 157–180. Sporer, S. L., Penrod, S. D., Read, J. D., & Cutler, B. L. (1995). Choosing, confidence, and accuracy: A meta-analysis of the confidence–accuracy relation in eyewitness identification studies. Psychological Bulletin, 118, 315–327. http://dx.doi.org/10.1037/ 0033-2909.118.3.315 Steblay, N. M. (1992). A meta-analytic review of the weapon focus effect. Law and Human Behavior, 16, 413–424. Tredoux, C. G. (1998). Statistical inference on measures of lineup fairness. Law and Human Behavior, 22, 217–237. Tryon, W. W. (2001). Evaluating statistical difference, equivalence, and indeterminacy using inferential confidence intervals: An integrated alternative method of conducting null hypothesis statistical tests. Psychological Methods, 6, 371–386. http://dx.doi.org/10.1037/1082-989X.6.4.371 Weber, N., & Brewer, N. (2004). Confidence–accuracy calibration in absolute and relative face recognition judgments. Journal of Experimental Psychology: Applied, 10, 156–172. http://dx.doi.org/10.1037/1076-898X.10.3.156 Weber, N., & Brewer, N. (2006). Positive versus negative face recognition decisions: Confidence, accuracy, and response latency. Applied Cognitive Psychology, 20, 17–31. http://dx.doi.org/ 10.1002/acp.1166 Wells, G. L., & Bradfield, A. L. (1998). “Good, you identified the suspect”: Feedback to eyewitnesses distorts their reports of the witnessing experience. Journal of Applied Psychology, 83, 360–376. Wells, G. L., & Murray, D. M. (1984). Eyewitness confidence. In G. L. Wells, & E. F. Loftus (Eds.), Eyewitness testimony: Psychological perspectives (pp. 155–170). New York, NY: Cambridge University Press. Wixted, J. T., & Mickes, L. (2012). The field of eyewitness memory should abandon probative value and embrace receiver operating

WEAPON CONFIDENCE ACCURACY

characteristic analysis. Perspectives on Psychological Science, 7, 275–278. Wixted, J. T., & Mickes, L. (2015). ROC analysis measures objective discriminability for any eyewitness identification procedure. Journal of Applied Research in Memory and Cognition, 4, 329–334. http://dx.doi.org/10.1016/j.jarmac.2015.08.007 Wixted, J. T., Mickes, L., Clark, S. E., Gronlund, S. D., & Roediger, H. L. (2015). Initial eyewitness confidence reliably predicts eyewitness identification accuracy. American Psychologist (in press).

92

Yaniv, I., Yates, J. F., & Smith, J. E. K. (1991). Measures of discrimination skill in probabilistic judgment. Psychological Bulletin, 110, 611–617. http://dx.doi.org/10.1037/0033-2909.110.3.611

Received 14 January 2016; received in revised form 30 March 2016; accepted 1 April 2016 Available online 26 May 2016