Skip to main content

Currently Skimming:

5 Applied Eyewitness Identification Research
Pages 71-102

The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.


From page 71...
... Eyewitness identification research resides in both the scientific literature and the law and justice-related scholarly literature. Although experiential, anecdotal, and some administrative records from law enforcement and the judiciary could contribute to a better understanding of eyewitness identification, the committee did not comprehensively review this more qualitative material.
From page 72...
... These materials included meta-analyses and systematic reviews and primary research in neuroscience, statistics, and eyewitness identification. This report also was informed by several early foundational papers and written comments from, and presentations to, the committee by representatives from science, law enforcement, state courts and government, private organizations, and other interested parties.
From page 73...
... In their examination of this body of literature, the committee examined the quality of the identified research and, where possible, worked to derive summary empirical generalizations related to variables of interest. Quantitative Syntheses of Eyewitness Identification Research The committee first evaluated the consistency of research findings across studies for system and estimator variables by studying published quantitative reviews of empirical research.
From page 74...
... In short, both systematic reviews and the studies they include need to be transparent and reproducible in order to best inform practice and policy decisions about eyewitness identification. The committee examined quantitative reviews that covered decades of research on both estimator variables (exposure duration,7 retention interval,8 stress,9 weapon focus,10 own-race bias,11 and own-age bias12)
From page 75...
... K Steblay et al., "Eyewitness Accuracy Rates In Sequential and Simultaneous Lineup Presentations: A Meta-Analytic Comparison," Law and Human Behavior 25(5)
From page 76...
... This question is, in principle, relevant to all system variables, but we address it first in the timely and controversial context of simultaneous versus sequential lineup presentations and in the role of eyewitness confidence judgments in evaluation of identification performance. This examination of lineup procedures and confidence reports is followed by a brief discussion of the effects on eyewitness performance of another important system variable: the extent and content of communications between the witness and the larger community (law enforcement, legal defense, the press, family and friends, etc.)
From page 77...
... , and the frequencies in each part of that table are the raw data used to evaluate performance on a binary classification task, such as eyewitness identification.27 There are many different performance measures that can be derived from data of this sort -- indeed, the fields of statistical classification and machine learning are replete with tools for the evaluation of binary classifiers.28 25  he T binary classifier in this context is defined as the witness operating under a specific set of conditions, such as lineup procedures. 26  lso termed "confusion matrix." A 27  he prevalence or "base-rate" -- the fraction of individuals in each category (guilty or in T nocent, in the eyewitness problem)
From page 78...
... Perhaps the simplest measure of binary classification performance is the ratio of hit rates (HR) to false alarm rates (FAR)
From page 79...
... lead to fewer false alarms without changing the frequency of hits, which would result in a higher diagnosticity ratio.34 More recent laboratory-based studies and meta-analyses typically show that sequential procedures (relative to simultaneous) are associated with a somewhat reduced hit rate accompanied by a larger reduction in the false alarm rate, thereby resulting in diagnosticity ratios higher than those yielded by simultaneous procedures.35 In other 31  .
From page 80...
... In light of these findings, many policy makers have advocated sequential procedures, and those procedures have been adopted by law enforcement in many jurisdictions. While policy decisions and practice have been influenced by the aforementioned studies, there are other criteria worthy of consideration when evaluating eyewitness performance.
From page 81...
... .a There are, of course, many other variables that will affect the outcome (e.g., levels of stress, attentional focus, potential rewards or expectations) , but all of these are believed to exert their influence over memory based classification decisions by affecting discriminability and/or response bias.
From page 82...
... , meaning he or she will have more hits and will make more false identifications. Differences in pick frequency can, and generally do, lead to differences in the ratio of hit rates to false alarm rates; all else being equal, the diagnosticity ratio will be higher for a conservative bias than for a liberal bias.37 In other words, simply by inducing a witness to adopt a more conservative bias, it is possible to increase the likelihood that an identified person is actually guilty.
From page 83...
... for simultaneous lineups is as high, or higher, than that for sequential lineups.41 In other words, when eyewitness identification performance is evaluated based on a criterion of bias-free discriminability, the results differ from those based on a single diagnosticity ratio, and they do so because the latter fails to account for response bias. Looking broadly at the many empirical studies that have used a single diagnosticity ratio to evaluate eyewitness performance, as well as the more recent findings using ROC analysis, it appears that the practical advantage of one lineup procedure over another depends to a large degree upon the performance criterion that one adopts.
From page 84...
... is a method from signal detection theory that enables one to distinguish the relative influences of discriminability and response bias on binary classification decisions. ROC analysis is performed by plotting the frequency of decisions that are hits (correctly detecting a target)
From page 85...
... in their responses. 1.0 Hit Rate 0 1.0 False Alarm Rate Receiver Operating Characteristic (ROC)
From page 86...
... 43 n reality, a more conservative bias may not always be beneficial, and whether it is or I not depends upon a number of factors that have an impact distinct from diagnostic accuracy and are difficult to quantify. All else being equal, the "best" response bias will be one that maximizes the "expected value" of the outcome (Green and Swets, Signal Detection Theory and Psychophysics; Swets, "ROC Analysis Applied to the Evaluation of Medical Imaging Techniques")
From page 87...
... See Appendix C 47  ee, e.g., Table 1 of Mickes, Flowe, and Wixted, "Receiver Operating Characteristic Anal S ysis of Eyewitness Memory," which summarizes confidence ratings, hit rates, false alarm rates, and diagnosticity ratios (HR/FAR)
From page 88...
... Both hit rates and false alarm rates declined steeply -- implying an increasingly conservative response bias -- as confidence levels increased. Diagnosticity ratios increased monotonically with increasing confidence.
From page 89...
... To illustrate this opportunity by example, we consider the following possibilities. It has been argued that a basic weakness of the existing ROC approach to binary classification performance results from the fact that, in principle 51  arlson C and Carlson, "An Evaluation of Lineup Presentation." Mickes, Flowe, and Wixted, "Receiver Operating Characteristic Analysis of Eyewitness Memory." 52  long the same lines, accuracy of discriminability measures derived from ROC studies A may be called into question when those studies do not take into account uncertainty in the data used to construct the ROC curves; see Appendix C
From page 90...
... As discussed above, the diagnosticity ratio is a critical piece of information in efforts to evaluate eyewitness performance. As for any binary classification, however, there is also information associated with a negative response, which is the predictive value of a classifier's assertion that a target is not present (in the eyewitness case, the witness' assertion of innocence)
From page 91...
... can be subjected to ROC analysis to account for the effects of response bias in the same manner as PPV-related measures (quantified as positive likelihood ratios, i.e., diagnosticity ratios) -- the ROC axes in the NPV case corresponding to 1-HR and 1-FAR.
From page 92...
... The committee's review revealed the need for further empirical research in individual studies and systematic reviews of research on these factors. The committee's review focused on the most-studied estimator variables: weapon focus, stress and fear, own-race bias, exposure, and retention interval.
From page 93...
... A 1992 analysis of weapon focus studies found that the presence of a weapon reduced both identification accuracy and feature accuracy (e.g., the eyewitness' ability to recall clothing, facial features, and more) .67 A more recent analysis of the weapon focus literature concluded that the presence of a weapon has an inconsistent effect on identification accuracy, in that larger effect sizes were observed in threatening scenarios than in non-threatening ones.68 As the retention interval increased, the weapon focus effect size decreased.
From page 94...
... may be significantly impaired.74 In the particular case of weapon focus, it may not be possible to sufficiently test the effects of stress and heightened stress in the laboratory because of limitations on human participant research that uses realistic and heightened threats. A meta-analysis of the effect of high stress on eyewitness 70  ee, S e.g., Fawcett et al., "Of Guns and Geese." 71  Deffenbacher et al., "A Meta-Analytic Review of the Effects of High Stress." 72  .
From page 95...
... The study also found that memories acquired during stressful events are highly vulnerable to modification by exposure to post-event misinformation, even in individuals whose level of training and experience might be considered relatively immune to such influences. Another recent study comparing the eyewitness accuracy of officers and citizens, concentrated on the effects of stress and weapon focus.82 The results of this study showed that officers were less stressed and aroused than 75  effenbacher D et al., "A Meta-Analytic Review of the Effects Of High Stress." It should be noted that the effect sizes for stress-induced support were small with wide confidence intervals, indicating considerable heterogeneity across studies.
From page 96...
... In eyewitness identification, own-race bias describes the phenomenon in which faces of people of races different from that of the eyewitness are harder to discriminate (and thus harder to identify accurately) than are faces of people of the same race as the eyewitness.83 In the laboratory, this effect is manifested by higher hit rates and lower false alarm rates (higher diagnosticity ratio)
From page 97...
... Until the scientific basis for these effects is better understood, great care may be warranted when constructing lineups in instances where the race of the suspect differs from that of the eyewitness. Exposure Duration Eyewitness identification researchers have long believed that exposure duration (e.g., time spent observing a perpetrator's face during a crime)
From page 98...
... Although, in general, it appears that longer retention intervals are associated with poorer eyewitness identification performance, the strength of this association appears to vary greatly across the circumstances of the initial encounter, identification procedures, and research method 93  .
From page 99...
... 99  . Sauer et al., "The Effect of Retention Interval on the Confidence–Accuracy Relationship J for Eyewitness Identification," Law and Human Behavior 34: 337–347 (2010)
From page 100...
... will elucidate those variables with meaningful influence on eyewitness performance, which can inform law enforcement practice of eyewitness identification procedures. To date, the eyewitness literature has evaluated procedures mostly in terms of a single diagnosticity ratio or an ROC curve; even if uncertainty is incorporated into the analysis, many other powerful tools for evaluating a "binary classifier" are worthy of consideration.100 When primary studies such as those described above are available in sufficient quantities, it is important that their results are synthesized using systematic reviews that conform to current best standards.101 These quantitative reviews would necessarily employ transparent, reproducible procedures for locating all relevant published and unpublished research; employ independent, duplicate procedures for selection of studies, extraction of data, and assessment of risk of bias; use meta-analytic procedures 100  astie, H Tibshirani, and Friedman, The Elements of Statistical Learning.
From page 101...
... can be used to further refine the research agenda in eyewitness identification research and to establish priorities for funding of additional primary research.


This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More information on Chapter Skim is available.