Victoria Beck and Chris Rose examine factors related to eyewitness identification following a staged theft. Study participants' misinformation was high but their confidence in the information provided was high. The authors identify several factors associated with misinformation.
There is a wealth of research on eyewitness accuracy dating back to the early 1900s, which has identified a variety of factors influencing eyewitness misinformation and misidentifications. However, this body of research has primarily utilized laboratory designs that do not reflect the reality of an eyewitness experiencing a criminal event. The current study expands research in this area by uniquely utilizing experiential realism (a staged theft in real time) to reevaluate some of the known factors related to eyewitness misidentification. After viewing a staged theft, study participants were randomly assigned to experimental conditions, and their ability to accurately identify the perpetrator was assessed. The primary results indicate that close to 69 percent of eyewitnesses provided inaccurate information, but more than half were confident they were accurate. Factors such as biased lineup instructions, co-witness misinformation, distance from the crime, and retention intervals influenced eyewitness accuracy. Ongoing consistency in empirical results across studies for eyewitness misidentifications, despite differences in research methodologies, should prompt national change in how eyewitness testimony is used in criminal cases.
Research has estimated that eyewitness testimony has been the sole or primary evidence of defendants’ guilt in 77,000 criminal trials each year in the United States (see Wells, et al., 1998). Indeed, eyewitness testimony is present more often in cases accepted for prosecution, when compared to cases rejected for prosecution (Flowe, et al., 2011). Such a heavy reliance on eyewitnesses by the courts is disconcerting, given that research on eyewitness accuracy has long highlighted the malleability of memory (Loftus, et al., 1978), and considering eyewitness misidentification is one of the leading causes of wrongful convictions (Garrett, 2008).
In this study we explore the malleability of eyewitness memory by examining how eyewitness memory can be influenced by and conform to what is said by others (e.g. co-witnesses, authority figures), even when that information is inaccurate (the misinformation effect). Although there is a wealth of research on memory conformity and the misinformation effect, existing studies have tended to employ research designs incorporating word parings, sequencing pictures of a crime, providing participants with information about what co-witnesses have said (Wright, et al., 2009), and/or crime videos (staged or real). Sporer (2008) argues that experimental procedures used to study eyewitness accuracy must strive to incorporate empirical realism (as opposed to laboratory manipulation), to improve external and ecological validity. Unfortunately, there is a dearth of research incorporating empirical realism when examining eyewitness accuracy and the few existing studies are contradictory. For example, Ihlebaek, et al. (2003) found participants who watched a video recording of a crime reported more details with higher accuracy, compared to a control group experiencing a simulated crime, however, Pozzulo, et al. (2008) did not find mode of exposure (video versus real time) to be significantly related to eyewitness accuracy across target present/absent lineups.
The primary goal of the current study was to address Sporer’s (2008) critique and provide additional insight into this area of research, by exploring several factors related to eyewitness accuracy through empirical realism. Using a simulated crime (a theft) in real-time the current study examined the relationship between eyewitness accuracy and: (1) the misinformation effect presented through post-event co-witness misinformation and biased lineup instructions for target-present and target-absent lineups; (2) length of time between witnessing a crime and being asked to identify the perpetrator of the crime (retention interval); (3) eyewitness confidence; and (4) distance from the stimulated crime. The following sub-sections of this literature review provide an overview of the research on the aforementioned variables to be included in the study.
One of the most noted factors exerting a substantial impact on eyewitness inaccuracy is the misinformation effect, which occurs when the memory is impaired due to sharing of inaccurate information (Loftus, 2005). An eyewitness may encounter misinformation through: (1) co-witnesses (Gabbert, et al., 2003; Gabbert, et al., 2004; Loftus, et al., 1978; Mori & Kishikawa, 2014; Patterson & Kemp 2006; Shaw, et al., 1997; Wright, et al., 2000;), or (2) leading statements made by people in authority (Loftus, 1979), such as police lineup administrators (Malpass & Devine 1981; Steblay, 1997; Warnick & Sanders, 1980;).
Studies focused on the contaminating influence of misinformation on eyewitness accuracy indicate that witnesses exposed to co-witness misinformation are likely to conform to that misinformation (Mori & Kishikawa, 2014; Shaw, et al., 1997; Wright, et al., 2000; Wade, et al., 2010), and make inaccurate identifications (Gabbert, et. al., 2003). When examining whether misinformation exerts more of an influence when introduced through a co-witness or through leading questions (such as might be presented by a police officer), Shaw, et al. (1997) found both produced a statistically significant effect on eyewitness misidentification. However, Patterson and Kemp (2006) found participants to be statistically significantly more likely to report post-event misinformation when it was introduced through a co-witness, compared to when the same misinformation was presented in the form of leading questions or media reports. The results in Gabbert, et al. (2004) indicate that when compared to control participants, experimental participants were significantly more likely to be influenced by misinformation introduced socially during verbal discussions, as opposed to non-social written misinformation.
Regardless of how misinformation is introduced to eyewitnesses, Kohnken and Mass (1988) argue that it is important for research on this topic to include target-absent (criminal absent) lineups to examine the least desirable type of identification – false selections. One of the first studies on pre-lineup police instructions was conducted by Warnick and Sanders (1980) and suggest that providing subjects with an unbiased line-up selection option of ‘don’t know’ decreases false identifications. In Malpass and Devine (1981), subjects viewed lineups with either the criminal present (target-present) in the lineup or with the criminal absent (target-absent) from the lineup. Additionally, subjects in this study received biased or unbiased pre-lineup instructions. The biased instructions led witnesses to believe that the criminal was in the lineup and did not provide for a ‘no choice’ option. The unbiased instructions did provide for a ‘no choice’ option and eyewitnesses were told that it is possible that the criminal may not be in the lineup. Malpass and Devine (1981) found subjects were more likely to choose when receiving biased instructions, especially under the target-absent condition. Under target-present lineups, errors were relatively low regardless of pre-lineup instructions (Malpass & Devine, 1981).
Kohnken (1985) also considered the influence of biased pre-lineup instructions on eyewitness accuracy when using target-present or target-absent lineups and receiving unbiased or biased pre-lineup instructions from a police officer. Contrary to results reported in Malpass and Devine (1981), the Kohnken (1985) study indicated that biased instructions increased the frequency of ‘don’t know’ responses for subjects, rather than increasing false positives (i.e., selecting the incorrect person). However, the subjects in Kohnken (1985) were unaware of being research participants, as they were not debriefed until after selecting a culprit from the lineup. Kohnken (1985) argues that undebriefed subjects may use a more strict decision criterion when selecting individuals from a police lineup. Similar to Malpass and Devine (1981) a meta-analytic review of research exploring the influence of police pre-lineup instructions on eyewitness accuracy indicated that unbiased lineup instructions are less likely to produce false-positives (Steblay, 1997).
In a review of studies exploring the misinformation effect, Loftus (2005) writes that, according to experts, misinformation is less likely to influence memory when the duration between the misinformation and culprit identification (i.e., retention interval) is shorter. In fact, research dating back to the early 20th Century (e.g., Ebbinghaus, 1908) has continuously noted problems with memory decay, and experts have continued to assert that there is a very rapid decrease in accuracy of memory over time, followed by a leveling-off effect (Kassin, et al., 1989; see also Penrod, et al., 1982). A meta-analysis of 53 studies further advanced this area of research by establishing a statistical consistent correlation between longer retention intervals and decreased memory for face recognition (Deffenbacher, et al, 2008).
Loftus (2005) maintains that eyewitness misinformation can “lead people to have very rich false memories. Once embraced, people can express these false memories with confidence and detail” (p. 365). Indeed, misled eyewitnesses can have as much confidence in their mistaken recognition judgements as non-misled eyewitnesses (Loftus, et al., 1989). While court procedures, such as cross-examination, are designed to help jurors and/or judges differentiate between a witness who is intentionally deceptive and one who is confidently truthful, those procedures do not assist in identifying a witness who is attempting to be truthful but is mistaken (Wells, et al., 1998). Thus, assessing the relationship between eyewitness confidence and memory accuracy is important, since the confidence of a witness may erroneously influence future events in the legal process (Wells et al., 2000).
Mudd and Govern (2004) used a sample of 27 study participants to explore the relationship between co-witness misinformation, retention intervals (immediate and two weeks), and found participants to be more confident in their misinformation over time. In a larger study with 232 adult participants, Paz-Alonso and Goodman (2008) discovered that, overall, participants were equally confident about misled and correct responses, and misinformed participants became more confident of their false memories over time. Early research has found a weak to non-existent relationship between confidence and memory accuracy (see Krug, 2007), with most experts agreeing that confidence is a limited predictor of memory accuracy (Simons and Chabris, 2011). For example, Goodwin, et al. (2017) found that a highly confident witness can create confidence conformity when sharing information among co-witnesses, even if that information is inaccurate (Goodwin, et al, 2017; Thorley & Kumar, 2016).
“It is a matter of common sense that a person is easier to recognize when close than when far away” (Loftus & Harley, 2005, p. 43). Indeed, Loftus and Harley (2005) found that after 25 feet, face recognition for people with normal vision diminishes and recognition drops to 0 at approximately 150 feet. In a study utilizing archival data from police lineups in 240,000 cases, Horry, et al. (2014) found that in corroborated cases, suspects were more likely to be identified when the perpetrator had been seen at a distance of less than 5 meters (approximately 16 feet).
To reiterate, the goal of the present study was to examine some of the known factors (misinformation effect, retention intervals, confidence, and distance) related to eyewitness accuracy through the lens of experimental realism to determine if our outcomes would differ from prior research. To employ more experimental realism, we paid an actor to steal a brief case placed on a table in a classroom during class time. Subsequent sections will discuss the study design, outcomes, limitations and conclusions.
After obtaining Institutional Review Board approval, subjects of the experiment witnessed a simulated crime, in real time. When first witnessed, the subjects in this experiment were unaware that the crime had been staged. As far as subjects were concerned, they had witnessed a male (wearing blue jeans, a heavy beige winter coat, and a yellow baseball hat) enter a door at the front of the classroom about five minutes after a regular class lecture had begun. The male confederate (i.e., research assistant) paused in the doorway for about five seconds until the professor asked if he needed any help. Immediately after being acknowledged by the professor, the confederate thief grabbed the professor’s personal belongings (either a brief case or a purse that had been placed on a table near the door) and ran out the door he had entered. To add to the realism, the professor yelled at the thief while she or he chased him out of the classroom and down the hallway. Due to the theater style classroom where each simulated crime was conducted, subjects had an unobstructed view of the simulated theft.
Upon returning to the classroom in an excited state (about 30 seconds after she or he had chased the thief out of the room), the professor asked the students (who were still unaware that the crime had been staged) if they had seen the culprit. In half of these simulated-crime experiments, the professor had been instructed to immediately say the following once they had returned to the classroom: Did you see that? Did you get a look at him? I noticed he had black hair. In the other half of these simulated-crime experiments, the professor, upon returning to the classroom, simply said: Did you see that? Did you get a look at him?
The phrase, I noticed he had black hair, was the co-witness misinformation that we planted in a little over half of the witnesses’ minds (n = 322). Whether or not a particular class would receive this co-witness misinformation was based on random assignment, which was determined by the flip of a coin. It should be noted that the confederate thief most assuredly did not have black hair; he was completely bald; however, he wore a baseball hat, which means his lack of hair may not have been immediately noticeable to the witnesses. Nonetheless, no one could have witnessed the thief having black hair.
After about a minute of chatter among the audience of eyewitnesses, the professor finally announced to the subjects that they did not actually witness a crime. The crime had been simulated as part of a larger experiment on eyewitness accuracy, and they were invited to continue to participate in the experiment if they provided their informed consent (to obtain IRB approval we were required to disclose this information to the subjects at that time). Once informed consent was acquired, subjects were then randomly assigned to one of several different experimental conditions that could vary by retention intervals (i.e., the amount of time they would have to wait before attempting to identify the thief: 1 hour, 2 days, or 7 days), and the way their eyewitness information would be collected (one of four possible scenarios in which the photo-lineup and instructions conveyed to the subjects varied).
The photo lineup contained six very dissimilar white males, with varying degrees of facial hair (the confederate had none), and two with short dark hair and the remaining four with little to no hair (the confederate was bald). In the first scenario, the subjects viewed a six-photo-lineup that did NOT include a picture of the confederate who had played the role of thief. Furthermore, the subjects in this first scenario were NOT warned of the fact that the thief may not appear in the lineup. In scenario number two, the subjects were given the same instructions (i.e., they were NOT warned that the thief may not appear in the lineup), but the six-photo-lineup that they viewed did include a picture of the thief. Those subjects randomly placed in scenario three viewed a six-photo-lineup in which the thief did NOT appear in the lineup, but they were warned that the actual thief may not appear. The remaining subjects were randomly placed in the fourth scenario in which the thief appeared in the lineup, and they were warned that the actual thief may not appear. All subjects, regardless of the scenario they found themselves in, were given the option to indicate that the thief does not appear in the photo-lineup, and an option that allowed them to indicate that they did not know whether the thief appeared. What varied from scenario to scenario was whether subjects were warned that the thief may not appear and whether he did appear in the lineup.
Figure 1 provides an image of the experimental design matrix that includes all the varying conditions that were investigated.
The numbers (1 through 24) presented in Figure 1 represent each possible experimental condition in which each subject could have been randomly placed. These 24 experimental conditions varied by: 1) retention intervals (1 hour, 2 days, or 7 days), 2) co-witness misinformation (i.e., a little over half of subjects were led to believe that the thief had black hair, n = 322, and a little less than half were not: n = 306), and 3) the scenarios in which the eyewitness information was collected (i.e., whether the thief appeared in the photo lineup as well as the type of pre-lineup instructions that subjects were given relating to whether the thief may or may not appear in the lineup). For example, in Figure 1, the experimental condition labeled 10 represents a scenario wherein subjects: 1) after witnessing the crime waited two days before viewing the photo lineup, 2) they heard misinformation about the thief, 3) the photo lineup they viewed did include a picture of the thief, and 4) they were NOT warned of the possibility that the thief may NOT appear in the lineup. In comparison, the condition labeled 13 involves a scenario where subjects also waited two days prior to viewing the photo lineup, but they did NOT hear a false description of the thief.
Furthermore, the photo lineup they viewed did NOT include a picture of the thief and they were NOT warned of this possibility.
We should also note that we included the subject distance from the spot of the crime, as well as subjects’ initial confidence in their ability to identify the thief (prior to subjects viewing the photo lineup). These variables are not included in the depicted experimental matrix, because we could not manipulate these variables during the experiment. Distance from the crime was simply determined by recording each subjects’ seat location. Primarily, distance was used as a control variable since someone’s proximity to a crime could influence their eyewitness accuracy over and above the conditions that we manipulated in this experiment.
With this experimental design, we explored the impact that these independent variables had on the likelihood that subjects could provide accurate eyewitness information. If the thief did not appear in the lineup, did the eyewitnesses accurately note this? If the thief did appear in the lineup, could the eyewitnesses accurately identify him? Since our ultimate dependent variable was dichotomous (i.e., subjects provided accurate eyewitness information after viewing the six-photo lineup or they did not), we used several dichotomous logistic regression analyses to assess the impact that each of these independent variables (co-witness misinformation, scenario experienced, retention interval, subject confidence in their ability to identify the thief, and distance) had on the odds that a subject could provide accurate eyewitness information.
Our dependent variable (accurate eyewitness information) was measured by asking all subjects to view a six-photo lineup and attempt to identify the thief. Each subject was provided eight response options. They could select the number that corresponded to the photo who they believed to be the thief (1 through 6). If they believed that the thief was NOT in the photo lineup they could indicate so (7). If unable to determine whether the thief was in the lineup (they simply did not know), they could indicate that they were unable to provide any useful information (8).
Based on those options, we created a dichotomy that would serve as our dependent variable. Any subject who provided inaccurate eyewitness information (they identified the wrong guy, or they said the thief was not in the lineup when he was) as well as any subject who indicated that they could not provide any useful information (8) was placed in the first category: accurate information NOT provided (coded as 0). Subjects were placed in the second category, accurate eyewitness information was provided (coded as 1), whenever a subject correctly picked the thief out of the lineup or correctly indicated that the subject did not appear in the lineup.
Ultimately, we entered five independent variables into a logistic regression equation designed to predict the odds subjects would provide accurate eyewitness information. These five predictor variables were: 1) distance from the spot of the simulated crime (measured in feet from a subjects seat to the spot of the crime), 2) co-witness misinformation (whether the professor presented misleading information about the thief), 3) the experimental scenario that the subject experienced (one through four), 4) the retention interval (whether subjects waited 1 hour, 2 days, or 7 days to provide their eyewitness information), and 5) subjects’ confidence in their ability to identify the thief (this confidence measure was taken prior to subjects’ viewing of the six-photo lineup). Table 1 provides the descriptive statistics for all variables in the logistic regression equation, and how we coded each attribute of each variable.
Table 1. Variables: Descriptive Statistics, Codes and Attributes, and Reference Categories used in the Logistic Regression Analyses.
Dependent Variable: Did subject provide accurate eyewitness information?
control variable (measured in feet from seat to spot of the crime)
9.84, min: 7, max 47
co-witness misinformation (professor provide false lead?)
*scenario experienced: (thief presence/line-up instructions):
(0) scenario 1
(1) scenario 2
(2) scenario 3
(3) scenario 4
(1) 1 hour
(2) 2 days
(0) 7 days
subject confident can ID:
An analytic goal of this study was to determine the odds that subjects would provide accurate eyewitness information about a thief committing a crime (scored as 1) based on how they scored on a variety of independent variables. This makes binary logistic regression analysis an attractive analytic procedure for our experimental design. Our dependent variable in this experiment was measured on a nominal scale and it was dichotomous (no = 0; yes = 1); binary logistic regression analysis is designed for exactly that type of dependent variable.
Furthermore, binary logistic regression analysis allows for both discrete and continuous variables to be entered into the predictive equation (four of our independent variables were discrete and the variable serving as a control variable, distance, was continuous). It also provides the added advantage in that binary logistic regression analysis will produce exponentiated beta coefficients (β) that express the odds that subjects will score a particular way on the dependent variable based on how they scored on a predictor variable. These exponentiated beta coefficients, for instance, allowed us to compare the odds that subjects who had a retention level of one hour would provide accurate information in comparison to subjects who had a retention level of seven days.
Finally, this analytic procedure is a form of regression analysis, which provides the ability to compare the relative impact of each independent variable on the dependent variable after partialling out the effects of the other variables in the equation. This was useful to control for the effects of distance on the other predictor variables’ impacts as we were conducting these experiments in a large lecture hall where some subjects would be very close to the spot of the crime and others would be much farther away.
Our sample was comprised of students attending a medium-sized Midwestern university (N = 628). The students who participated in this experiment were enrolled in either introductory criminal justice or philosophy courses, which were open to all university students who were seeking general education requirements. Students did not have to be criminal justice or philosophy majors to enroll in these courses. There was a close to even split among gender, with most students classifying their gender as male (51.9%) and the remaining classifying their gender as female (48.1%).
The average age of our subjects was 19.89 years (standard deviation = 2.29) with a median age of 19, and a minimum age of 18 and a maximum age of 45. Most of the students were of traditional college age from 18 to 22 years. It should also be noted that much of the sample (n = 588) indicated that they had never been asked by a criminal justice agency to identify a suspect in a photo-lineup, and most indicated that they had never witnessed anyone commit the crime of theft prior to this experiment (n = 409). While most subjects reported their major to be criminal justice (n = 338), the remaining subjects indicated a vast array of majors in which all colleges at the university were represented.
Most students classified their race as Caucasian (non-Latino) (n = 498). This over-representation of Caucasians was not surprising due to the regional nature of this Midwestern university and the surrounding population. The next largest racial category was Caucasian (Latino) (n = 27), which was followed by African Americans (n = 26) and Asian or Pacific Islanders (n = 26). Our sample also included a small number of students who classified their race as American Indian, Alaskan Native, or Aleut (n = 5), as well as several students (n = 15) who classified themselves as both Caucasian and African American. A total of 31 students did not provide a description of their race.
In Table 2, we show the exponentiated beta coefficients (β) derived from five logistic regression models. Each of these models were designed to predict the odds that subjects would provide accurate eyewitness information (after viewing the six-photo-lineup) based on how they scored on the independent variables (distance, co-witness misinformation, experimental scenario, retention interval, and subject confidence in their ability to identify the thief).
Table 2. Estimated Odds Ratio of Dichotomous Logistic Regression Analyses: Predicting Accurate Eyewitness Information (0 = accurate information not provided; 1 = accurate information provided)
Subject Confident Can ID
*p < .05; **p < .01; ***p < .001
Since distance is measured in feet, we have not interpreted it in terms of a reference category as we do with all other independent variables. Instead, we illustrate how much the odds of providing accurate eyewitness information diminished with a single increasing unit change in distance (a single unit change was equal to 5 feet; the distance between ascending rows in the room). Our results show that as subject distance from the crime increased by about 5 feet (a single unit change), their odds of providing accurate eyewitness information diminished by 5 percent (βmodel 5 = .95; b = -.056, p < .001). As a result, a subject that was 25 feet away from the simulated crime was about 5 percent less likely to provide accurate eyewitness information than a subject who was 20 feet away. The impact of distance was stable across all five of the logistic regression models with relatively little change in impact from model to model. As such, it is important variable to consider and/or control for when researching eyewitness accuracy.
When subjects were provided incorrect information about the thief by another witness (i.e., I noticed he had black hair.), they were about 34% less likely to provide accurate eyewitness information when asked to do so (βmodel 5 =.66; b = -.404, p < .05). Once the impact of this independent variable became significant, it remained relatively stable throughout the remaining models (βmodel 3 = .64; βmodel 4 = .65; βmodel 5 = .66).
For the reference category, we used the scenario in which the thief did NOT appear in the photo-lineup and during photo-lineup instructions subjects were NOT warned of this possibility (scenario 1). This scenario was chosen as the reference category because this is where subjects performed the worst. Of all subjects who experienced this condition (n = 154) only 16.2% (n = 25) provided accurate eyewitness information. If the thief did not appear in the photo-lineup and subjects were not warned of that possibility, it seemed to set them up to fail at providing accurate eyewitness information.
In comparison to our reference category, subjects in scenario two, where the thief appeared in the photo-lineup and during photo-lineup instructions subjects were still NOT warned of that possibility, subjects fared better. They were just over 4 and half times more likely to provide accurate eyewitness information than those in the reference category (βmodel 5 = 4.61; b = 1.53, p < .001). Subjects in scenario three (the thief did NOT appear in the photo-lineup and during instructions subjects were warned of this possibility) were just over 3 and half times more likely to provide accurate eyewitness information than those in the reference category (βmodel 5 = 3.81; b = 1.34, p < .001). Finally, subjects from the scenario in which thief appeared in the lineup and a warning was provided that the thief may not appear (scenario 4) were about 3 and half times more likely to provide accurate eyewitness information than reference category subjects (βmodel 5 = 3.56; b = 1.27, p < .001).
With retention interval, we used the category in which subjects performed the worst at providing accurate eyewitness information as the reference category (7 days). Subjects with a retention interval of one hour were about twice as likely to provide accurate eyewitness information than those subjects in the reference category (βmodel 5 = 2.01; b = .70, p < .01), and those with a retention interval of two days were over two and a half times more likely to provide accurate eyewitness information than those in the reference category (βmodel 5 = 2.72; b = 1.00, p < .001). Here, we should note that no significant difference in the proportion of subjects providing accurate eyewitness information was found between those subjects who experienced a one hour retention interval and those experiencing a two day retention interval (χ2 = 2.5, 1 df, p > .05). As such, while memory did eventually begin to falter, it did not significantly begin diminishing subject ability to provide accurate eyewitness information until after, at least, 2 days expired.
While most subjects (67.5%) expressed confidence in their ability to accurately identify the thief prior to viewing the lineup, a much lower percentage (31.4%) would provide accurate eyewitness information once they were given the opportunity to view the photo-lineup. While subjects seem to have been confident in their abilities to produce accurate results, it seems their abilities to do so may have been hampered. More than two-thirds of them (68.6%) would NOT provide accurate eyewitness information. They did, however, fare better when retention intervals were shorter. Thirty-three percent of those subjects who only waited an hour (n = 62) would be able to provide accurate eyewitness information; 41 percent (n = 62) were successful after waiting two days, but only 23 percent (n = 50) could be successful after seven days.
When subject confidence in their ability to identify the thief was entered into the final equation, it was found to be significant (p < .001), but not the way one might hope. Subjects who felt confident that they would be able to identify the thief when they were given the opportunity did not have higher odds of providing accurate eyewitness information. Instead, subjects who reported that they were confident that they could identify the thief were 56 percent less likely to provide accurate eyewitness information (βmodel 5 = .44; b = -.82, p < .001). Our subjects in this study seemed to be overconfident in their ability to identify the witness.
The results of our five logistic regression models show that distance, co-witness misinformation, scenario experienced by subject, retention interval, and subject confidence were all significant contributors in predicting whether subjects could provide accurate eyewitness information about a thief who perpetrated a simulated crime that subjects believed to be real.
For over a century researchers have highlighted the fallibility of eyewitness accuracy. Nonetheless, it has been implied that a lack of experimental realism (Sporer, 2008) in this area of research may be problematic and could, for example, result in overestimating eyewitness accuracy (Ihlebaek, et al., 2003).The primary goal of the current study was to employ experimental realism (simulate a theft) when examining some factors known to influence eyewitness accuracy, such as: co-witness misinformation, leading statements for target-present/absent lineups, witness confidence, retention intervals, and distance from the crime. To the best of our knowledge, our study is the first to combine experimental realism with random assignment to 24 experimental conditions and a large sample size. Notwithstanding the unique approach, the research results in the current study were consistent with prior research and continue to provide robust support for the focal variables.
Despite being in small, tiered lecture rooms with unobstructed views, the results from the current study indicated that close to 69 percent of the eyewitnesses provided inaccurate information. Further, inaccurate identification significantly increased with distance from the crime, which is consistent with the research results in Loftus and Harley (2005) and Horry, et al. (2014). The results in this study related to the misinformation effect were also consistent with prior research (e.g., Gabbert, et al., 2003; Gabbert, et al., 2004; Loftus, et al., 1978; Mori & Kishikawa, 2014; Patterson & Kemp 2006; Shaw, et al., 1997; Wright, et al., 2000). In the current study, when eyewitnesses were provided with misinformation about the fictitious thief from a co-witness, they were approximately 33 percent more likely to provide inaccurate descriptions of the thief and this effect became a significant and stable predictor of eyewitness inaccuracy. Nonetheless, more than half of the subjects were confident they had provided accurate information. In reality, a significantly (p < .001) higher percentage of eyewitnesses reporting that they were confident in their ability to identify the culprit failed to provide accurate information, in comparison to the percentage of confident subjects who could provide accurate information, which supports results from prior studies on pre-lineup confidence (Goodwin, et al. 2017; Mudd & Govern, 2004; Paz-Alonso & Goodman, 2008).
In the current study, eyewitness misidentification was particularly problematic for witnesses who were not warned the thief may not be in the photo lineup, and the thief was not in the lineup, consistent with the results in Malpass and Devine (1981). Only 16.2 percent of subjects provided accurate eyewitness information, when experiencing the scenario in which the thief did not appear in the photo-lineup and subjects were not warned of this possibility. When reviewing photo lineups of potential culprits, increasing delays between the fictitious theft and viewing the lineup resulted in a statistically significant (p < .001) reduction in accuracy after a two-day delay, which is consistent with prior research noting a rapid memory decay over time for eyewitness accuracy (Deffenbacher, et al., 2008; Kassin, et al., 1989; Penrod, et al., 1982). Finally, intuitively, and consistent with prior research (Horry, et al., 2014; Loftus & Harley, 2005) increased distances between the fictitious theft and the eyewitness significantly (p < .001) reduced identification accuracy.
Although the current research approach of employing experimental realism resulted in consistent findings with prior research for the focal variables, there are a few notable concerns. As discussed in Kohnken (1985), subject debriefing prior to lineup identifications may be problematic. In the current study, the Internal Review Board required the deception inherent in the simulated crime be made apparent to research subjects immediately after the theft, which may reduce the influence of experimental realism. Informing subjects of the research project immediately after the theft, but prior to the photo lineup identification process, may have negatively influenced the validity of the experiment by increasing false positives from lineup selections. When subjects are aware that selecting a person from the photo lineup will not, under an experiment, result in a negative outcome for the person they select, subjects may be less cautious in making such selections. For example, in one of the few studies employing experimental realism, conducted in Germany, subjects were not debriefed until after the photo lineup and the researcher found that biased instructions increased the number of “don’t knows” instead of increasing false positives (Kohnken, 1985). According to Kohnken (1985), as previously mentioned, subjects who are not debriefed may use a more strict decision criterion when selecting someone from a police lineup. Thus, future research should strive to maintain experimental realism until after the lineup identification, with experimental debriefing occurring at the end of the experiment. Perhaps if the study is conducted within a few hours on one day, with subject debriefing occurring at the end of the experiment an Internal Review Board may allow the experimental deception to continue until after the lineup. Although this approach would, of course, eliminate the simultaneous exploration of the influence of retention intervals on eyewitness accuracy.
Another notable concern focuses on the assessment of confidence and eyewitness accuracy. The current study assessed witnesses’ confidence in their ability to accurately select the thief from a lineup prior to viewing the lineup (predictive confidence) and found the relationship to be weak, which was consistent with prior research on this variable (Goodwin, et al. 2017; Mudd & Govern, 2004; Paz-Alonso & Goodman, 2008). However, according to more recent research by Nguyen, et al. (2018), evaluating the confidence of an eyewitness immediately after the viewing of a suspect lineup (postdictive confidence) may provide a more reliable assessment of confidence and a stronger correlate of eyewitness accuracy. Of course, a postdictive measure of confidence would be even more meaningful if subject debriefing takes place at the end of the experiment. Thus, future research utilizing experimental realism should strive to include a measure of eyewitness postdictive confidence. It may also be useful for future research to provide a comparison of predictive and postdictive confidence measures, using experimental realism, for comparative purposes.
While using student populations in eyewitness studies is a common practice, future research should endeavor to include nonstudent populations to improve generalizability. The importance of diversity in age for eyewitness research has been noted. In a meta-analysis of 91 studies, Fitzgerald and Price (2015) determined young adults (age range 19 –28) have a lower choosing rate in lineups, a higher identification accuracy in target present lineups, and a higher correct rejection rate in target absent line-ups, when compared with children (ages 4 –17) and older adults (ages 45–77). Although it might be difficult to obtain approval to specifically target children (who are identified as a protected group by Internal Review Boards) or an elderly population in research utilizing experimental realism, due to potential trauma caused by witnessing a theft. Notwithstanding the foregoing, the results in Fitzgerald and Price (2015) also suggest college-aged students may provide the most robust outcomes when examining factors influencing eyewitness misidentifications, as that age group appears to make less errors in identification when compared to other age groups.
Moreover, three-quarters of the research subjects in the current study were Caucasian (non-Latino). While reflective of the student population from which it was drawn, the lack of racial diversity in the sample, combined with the fact that the experimental design yielded 24 experimental conditions, resulted in not enough cases falling into any one racial category to provide a useful variable measuring race (other than Caucasian non-Latino). Consequently, it was not possible to determine if eyewitness accuracy differed across racial categories in this study, limiting generalizability. To be sure, racial diversity in eyewitness accuracy studies is an important consideration as prior research has established a cross-race effect on eyewitness accuracy, whereby same-race faces are recognized more accurately than cross-race faces (Brigham, et al. 1982; Brigham & Malpass, 1985; Chiroro & Valentine, 1995; Platz & Hosch, 1988). Additionally, Doyle and Bassil (2001) suggest that an own race bias may exist, with Whites applying a more lenient criteria when identifying Blacks. It should be noted, however, that while most of the subjects in this study were White, so was the thief which may minimize the limitation caused by a lack of a diverse sample for this study. Moreover, research results based on data from police lineups suggest that witnesses are likely to be more accurate in lineup identifications when the culprit is white (Valentine, et al. 2003). Nonetheless, future research in this area should strive to include a more diverse population to address the cross-race effect and own race bias. Short of having a diverse sample, future research should consider varying the race of the culprit to produce a cross-race effect.
Wrongful convictions of innocent people due to eyewitness misidentifications are one of the more egregious aspects of the criminal justice system, particularly in death penalty cases. The focus of the current study was to examine the known factors related to eyewitness misidentifications through the underutilized lens of empirical realism. The results served to provide robust support for existing research and suggest mode of experimental exposure (video of theft versus a simulated theft) may not matter. For decades research has demonstrated, through various methodologies, that there are significant reasons to be concerned about the use of eyewitnesses in criminal cases. Researchers have also suggested a variety of criminal justice reforms to safeguard against misidentifications (e.g., Cutler, 2013; Findley, 2016). Some suggested reforms have been adopted at the state-level in some states, such as jury instructions to help jurors evaluate eyewitness testimony (New Jersey v. Henderson, 2011), and allowing for the use of expert testimony on eyewitness fallibility (Oregon v. Lawson, 2012). Some reforms emanating out of research have been implemented at the local level, within police departments, such as: double-blind lineups, and unbiased lineup instructions (e.g., culprit may or may not be in the lineup). Nonetheless, to truly tackle this barrier to justice and create equality in the application of justice, it is crucial eyewitness misidentification be addressed through the United States Supreme Court as a due process issue under the Fourteenth Amendment. Unfortunately, the Court has afforded scant due process protection against eyewitness misidentification.
In Manson v. Brathwaite (1977, p 109-117), the United States Supreme Court held “The Due Process Clause of the Fourteenth Amendment does not compel the exclusion of the identification evidence.” Instead, the Court emphasized that “reliability is the linchpin in determining the admissibility of identification testimony” and instructed judges to: (1) examine whether the identification procedures were unnecessarily suggestive; and (2) assess whether identification is reliable based on factors drawn from earlier judicial rulings (not scientific evidence). In other words, the Court is interested in witness veracity regarding reliability, not witness accuracy. Our research results make clear that witness veracity of reliability – correctness in identification – is not equal with actual correctness in identification. Over 30 years later, Perry v. New Hampshire (2012) sought to increase protections by excluding eyewitness testimony when witnesses identify perpetrators under suggestive circumstances. The American Psychological Association filed an Amicus Curiae brief in Perry v. New Hampshire (2012) highlighting decades of eyewitness scientific research and explaining: “Controlled experiments as well as studies of actual identifications have consistently found that the rate of incorrect identifications is approximately 33 percent” (p. 3). In their brief, the APA argued that due process protections should not be limited to faulty eyewitness identification procedures caused by state actors. Unfortunately, the Court rejected the argument for an expansion of rules to protect defendants from eyewitness misidentifications.
Evolution in criminal justice practices and advances in research techniques encourage ongoing research on eyewitness identification to develop an ever more discrete understanding of factors influencing and the issues surrounding misidentifications. It is incumbent upon researchers to continue to amass evidence to aid in a social justice push for due process protections against eyewitness misidentifications, and lobby for changes in court proceedings where eyewitness are included. For example, national changes should include, at a minimum: jury instructions on factors related to eyewitness misidentification; granting expert witness testimony regarding eyewitness identification when requested; ensuring the existence of corroborating real evidence in conjunction with eyewitness evidence, and/or the exclusion of eyewitness testimony under certain conditions (e,g., suggestive tactics or environment, distance from event over 16 feet, retention intervals longer than 2 days, cross-race identifications). Total dependence on cross examination and reliance on jury determination of witness reliability and credibility should be reconsidered by the judiciary, both on the trial and appellate levels. The consequence for those who have been misidentified (e.g., Kalif Browder, Ronald Cotton, Troy Davis) in the criminal justice system are too unbearable for the researchers to rest on their laurels.
Contact Author: Victoria Beck, Ph. D., Professor of Criminal Justice, Department of Criminal Justice, University of Wisconsin Oshkosh Algoma Blvd., Oshkosh, WI 54901, [email protected]
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(1), 11-30.
Brief for the American Psychological Association as Amicus Curiae, Perry v. New Hampshire, 565 U.S. 228, (2011). https://www.apa.org/about/offices/ogc/amicus/new-hampshire.pdf
Brigham, J. C., Maass, A., Snyder, L. D., & Spaulding, K. (1982). Accuracy of eyewitness identifications in a field setting. Journal of Personality and Social Psychology, 42, 673- 681.
Brigham, J. C., & Malpass, R. S. (1985). The role of experience and contact in the recognition of faces of own-and other-race persons. Journal of Social Issues, 41, 139-155.
Chiroro, P., & Valentine, T. (1995). An investigation of the contact hypothesis of the own-race bias in face recognition. The Quarterly Journal of Experimental Psychology, 48A, 879-894.
Couch, R. (2013). A model for fixing identification evidence after Perry v. New Hampshire. Michigan Law Review, III, 1535-1546.
Cutler, B.L. (Ed.) (2013). Reform of eyewitness identification procedures. American Psychological Association.
Doyle, J.M. & Bassil, C. (2001). Discounting the error costs: Cross-racial false alarms in the culture of contemporary criminal justice. Psychology, Public Policy, and Law, 7(1), 253-262.
Deffenbacher, K.A., Bornstein, B.H., McGorty, E.K. & Penrod, S.D. (2008). Forgetting the once-seen face: Estimating the strength of an eyewitness’s memory representation. Journal of Experimental Psychology Applied 14(2), 139-150.
Ebbinghaus, H. (1908). Psychology: An elementary textbook. New York: Arno Press.
Findley, K.A. (2016). Implementing the lessons from wrongful convictions: An empirical analysis of eyewitness identification reform strategies. Missouri Law Review 81(2), 1-77.
Fitzgerald, R. J., & Price, H. L. (2015). Eyewitness identification across the life span: A meta-analysis of age differences. Psychological Bulletin, 141(6), 1228–1265.
Flowe, H.D., Mehta, A. & Ebbesen, E.B. (2011). The role of eyewitness identification evidence in felony case dispositions. Psychology, Public Policy & Law, 17(1), 140-159.
Gabbert, F., Memon, A., and Allan, K. (2003). Memory conformity: Can eyewitnesses influence each other’s memories for an event? Applied Cognitive Psychology, 17, 533–543.
Gabbert, F., Memon, A., Allan, K., and Wright, D.B (2004). Say it to my face: Examining the effects of socially encountered misinformation. Legal and Criminological Psychology, 9 (2), 215–227.
Garrett, B. L. (2008). Judging innocence. Columbia Law Review, 108, 55–142.
Goodwin, K.A., Hannah, P.J., Nicholl, M.C., Ferri, J.M. (2017). The confident co-witness: The effects of misinformation on memory after collaborative discussion. Applied Cognitive Psychology, 31, 225-235.
Horry, R. Halford, P. Brewer, N., Milne, R. & Bull, R. (2014). Archival analyses of eyewitness identification test outcomes: What can they tell us about eyewitness memory? Law and Human Behavior 38 (1), 94 –108.
Ihlebaek, C. Love, T. Eilertsen, D.E. & Magnussen, S. (2003). Memory for a staged criminal event witnessed live and on video. Memory 11(3), 319-327.
Kassin, S.M., Ellsworth, P.C., & Smith, V.L. (1989). The “general acceptance" of psychological research on eyewitness testimony: A survey of the experts. American Psychologist, 44, 1089-1098.
Kohnken, G. (1985). Effects of suggestive instructions on the identification performance of eyewitnesses in simulated serious situations. Journal of Law and Experimental and Applied Psychology, 32, 281- 300.
Kohnken, G. and Maass, A. (1988). Eyewitness testimony: False alarms on biased instructions? Journal of Applied Psychology, 73 (3), 363-370.
Krug, K. (2007). The relationship between confidence and accuracy: Current thoughts of the literature and a new area of research. Applied Psychology in Criminal Justice 3(1), 7-41.
Loftus, E. F. (1975). Leading questions and the eyewitness report. Cognitive Psychology, 7, 560-572.
Loftus, E. F. (1979). Eyewitness testimony. Cambridge, MA: Harvard University Press.
Loftus, E.F. (2005). Planting misinformation in the human mind: A 30-year investigation of the malleability of memory. Learning & Memory 12, 361-366.
Loftus, E. F., Donders, K., Hoffman, H. G., & Schooler, J. W. (1989). Creating new memories that are quickly accessed and confidently held. Memory & Cognition, 17, 607–616.
Loftus, G. & Harley, E.M. (2005). Why is it easier to identify someone close than far away? Psychonomic Bulletin & Review, 12(1), 43-65.
Loftus, E.F., Miller, D.G., & Burns, H.J. (1978). Semantic integration of verbal information into visual memory. Journal of Experimental Psychology: Human Learning, 4 (1978), 19–31.
Malpass, R.S. & Devine, P.G. (1981). Eyewitness identification: Lineup instructions and the absence of the offender. Journal of Applied Psychology, 66(4), 482-489.
Manson v. Brathwaite, 432 U.S. 98 (1977). https://supreme.justia.com/cases/federal/us/432/98/.
Mori, K. & Kishikawa, T. (2014). Co-witness auditory memory conformity following discussion: A misinformation paradigm. Perceptual & Motor Skills: Perception 118(2), 533-547.
Mudd, K. & Govern, J.M. (2004). Conformity to misinformation and time delay negatively affect eyewitness confidence and accuracy. North American Journal of Psychology 6(2), 227-238.
Nguyen, T.B, Abed, E. & Pezdek, K. (2018). Postdictive confidence (but not predictive confidence) predicts eyewitness memory accuracy. Cognitive Research: Principles and Implications 3(32). https://doi.org/10.1186/s41235-018-0125-4.
Patterson, H.M. & Kemp, R.I. (2006). Comparing methods of encountering post-event information: The power of co-witness suggestion. Applied Cognitive Psychology 20, 1083– ISSN 0031-5125.
Paz-Alonso, PM. & Goodman, G.S. (2008). Trauma and memory: Effects of post-event misinformation, retrieval order, and retention interval. Memory 16(1), 58-75.
Penrod, S., Loftus, E. F., & Winkler, 1. (1982). Reliability of eyewitness testimony – A psychological perspective. In N. L. Kerr and R.M Bray, ed., Psychology of the courtroom, Cambridge, MA: Academic Press, 119-168.
Perry v. New Hampshire, 565 U.S. 228, (2011).
Platz, S. J., & Hosch, H. M. (1988). Cross-racial/ethnic eyewitness identification: A field study. Journal of Applied Social Psychology, 18, 972-984.
Pozzulo, J.D., Crescini, C. Panton, T. (2008). Does methodology matter in eyewitness identification research? The effect of live versus video exposure on eyewitness identification accuracy. International Journal of Law and Psychiatry 31(5), 430-437.
Shaw, J.S., 3rd, Garven, S. and Wood, J.M. (1997). Co-witness information can have immediate effects on eyewitness memory reports. Law and Human Behavior, 21(5), 503-523.
Simons, D.J. and Chabris, C.F. (2011). What people believe about how memory works: A representative survey of the U.S. population. PLoS One, 6 (8), 1-7.
Sporer, S.L. (2008). Lessons from the origins of eyewitness testimony research in Europe. Applied Cognitive Psychology 22(6), 737-757.
State of New Jersey v. Larry R. Henderson, 27 A.3d 872 (2011).
State of Oregon v. Samuel Adam Lawson, 291 P.3d 673, (2012).
Steblay, N.M. (1997). Social influence in eyewitness recall: A meta-analytic review of lineup instruction effects. Law and Human Behavior 21(3), 283-297.
Thorley, C. & Kumar, D (2017). Eyewitness susceptibility to co-witness misinformation is influenced by co-witness confidence and own self-confidence. Psychology, Crime and Law, 23(4), 342 – 360.
Valentine, T., Pickering, A. & Darling, S. (2003). Characteristics of eyewitness identification that predict the outcome of real lineups. Applied Cognitive Psychology 17, 969-993.
Vallas, G. (2011). A survey of federal and state standards for the admission of expert testimony on the reliability of eyewitnesses. American Journal of Criminal Law 39(1), 97-146.
Wade, K.A., Green, S.L. & Nash, R.A. (2010). Can fabricated evidence induce false eyewitness testimony? Applied Cognitive Psychology 24, 899-908.
Warnick, D. H. & Sanders, G.S. (1980). Why do eyewitnesses make so many mistakes? Journal of Applied Social Psychology 10(4), 362– 366.
Wells, G.L., Small, M., Penrod, S., Malpass, R.S., Fulero, S.M. & Brimacombe, C.A.E. (1998). Eyewitness identification procedures: Recommendations for lineups and photospreads. Law and Human Behavior, 22(6), 1-39.
Wise, R.A., Gong, X., Safer, M.A. & Lee, Y. (2010). A comparison of Chinese judges’ and US judges’ knowledge and beliefs about eyewitness testimony. Psychology Crime & Law 16(8), 695-713.
Wright, D. B., Memon, A., Skagerberg, E. M., & Gabbert, F. (2009). When eyewitnesses talk. Current Directions in Psychological Science, 18, 174–178.
Wright, D.B., Self, G., and Justice, C. (2000). Memory conformity: Exploring misinformation effects when presented by another person. British Journal of Psychology, 91 (2), 189-202.