Linking Eyewitness Recognition Memory to Identification Accuracy using Item Response Theory

Linking Eyewitness Recognition Memory to Identification Accuracy using Item Response Theory In legal proceedings, eyewitness testimony often serves as crucial evidence, particularly when objective forms of evidence, such as DNA or fingerprints, are unavailable. Although human memory is in general reliable (Wixted & Mickes, 2022), it can also be fallible and errors in eyewitness testimony remain a leading cause of wrongful convictions (National Registry of Exonerations, 2025). Consequently, researchers have long investigated the conditions under which eyewitness testimony is reliable and the factors that can compromise it (e.g., Lindsay et al., 2025). Broadly, eyewitness research has focused on two related but distinct aspects of eyewitness performance: recalling experienced events (eyewitness memory) and identifying a perpetrator from a lineup (eyewitness identification). While extensive work has examined how factors such as interviewing techniques, stress, or intoxication influence these processes (e.g., Jores et al., 2019; Shields et al., 2017), relatively little attention has been given to the relationship between eyewitness memory and identification accuracy. To meaningfully examine such a relationship, it is critical that the measures used truly capture the constructs they are intended to assess. However, the psychometric properties—and the associated measurement error—of eyewitness memory measures in experimental studies have rarely been evaluated. The present study addresses these gaps by assessing the psychometric properties of eyewitness memory measures and examining the relation of these memory measures to eyewitness identification performance. The Relationship Between Eyewitness Memory and Identification An assumption in eyewitness research is that individuals with better memory for an event or perpetrator should also make more accurate identifications. However, empirical evidence for this relationship is scarce, and the research that has been conducted shows mixed results. One of the earliest studies to address this issue was conducted by Wells and Leippe (1981). In their experiment, participants witnessed a staged theft, later attempted to identify the perpetrator from a target-present lineup, and completed a cued-recall test assessing memory for 11 peripheral event details (e.g., the number of chairs in the room). Surprisingly, participants who made correct identifications remembered fewer peripheral details than those who made incorrect identifications—suggesting that stronger eyewitness memory for peripheral details is related to lower eyewitness identification accuracy. Further research suggests that the link between eyewitness memory and identification accuracy is weak. For instance, Pozzulo and colleagues (2009) found that participants who made correct lineup decisions (i.e., correct identifications or correct rejections) did not freely recall statistically significantly more details, or more accurate details, about the crime or perpetrator compared with participants who made incorrect lineup decisions. Similarly, Sheahan et al. (2017) examined the relationship between descriptor recall (i.e., recall of details describing the perpetrator) and identification accuracy and found that participants who correctly identified the perpetrator in target-present lineups did not recall statistically significantly more descriptors than those who made incorrect identifications. One potential explanation for the weak or inconsistent relationship between eyewitness memory and identification accuracy could be that recall and recognition represent distinct types of memory (Tulving, 2011), each relying on partly different cognitive mechanisms. Specifically, it has been proposed that recall mainly relies on conscious recollection whereas recognition is based on familiarity and recollection (Mandler, 1980). Because eyewitness identification is a form of recognition memory, measures of recognition-based eyewitness memory may be more theoretically aligned with identification performance than recall measures. The previous research presented above exclusively used recall measures to examine eyewitness memory and its relation to lineup identification, and this is one explanation for the mixed and weak findings regarding the link between eyewitness memory and identification. Indeed, to our knowledge, no prior research has directly examined whether performance on a recognition-memory task for the mock-crime event is related to the eyewitness identification accuracy for the corresponding perpetrator(s). This is relevant as witnesses might be confronted with recognition based questions during an interview at the police and subsequently presented with a lineup. The present study thus examines the relationship between recognition-based eyewitness memory and identification performance. Another explanation for the mixed or weak findings regarding the relationship between eyewitness memory and identification is that methodological and inferential practices common in eyewitness research such as over-reliance on null-hypothesis significance testing, insufficient statistical power, limited considerations of effect sizes, and deficient assessment of psychometric properties may obscure meaningful associations between different memory measures and identification outcomes. Research practices in eyewitness memory and identification Despite a large body of research investigating factors that improve, undermine, or predict eyewitness memory and identification, recent reviews have highlighted several methodological and statistical issues in this literature that may limit the interpretation of eyewitness memory findings (Riesthuis & Otgaar, 2025a, 2025b; Riesthuis, 2025). For example, studies often rely only on statistical significance without considering practical or theoretical relevance (Riesthuis & Otgaar, 2025a), report incorrect or irreproducible power analyses (Riesthuis, 2025), and provide limited interpretation of effect sizes (Eastwood et al., 2025; Riesthuis & Otgaar, 2025a, 2025b). Such practices hinder meaningful interpretation and reduce the practical relevance of reported findings. These issues become especially problematic in the case of null findings, as traditional null-hypothesis significance testing (NHST) cannot provide evidence for the absence of an effect; tests of equivalence are required for such claims (Lakens et al., 2018). A nonsignificant result merely indicates insufficient data to conclude that the observed difference would be unlikely under the null hypothesis (Murphy et al., 2024). This limitation has real consequences: in Pozzulo et al. (2009), there was ‘no evidence’ that witnesses who made correct lineup decisions recalled more information. Yet, differences in the number of details recalled by participants who made correct versus incorrect lineup decisions were sometimes as large as two details—substantially larger than a commonly proposed smallest effect size of interest for eyewitness memory studies of one detail (e.g., Otgaar et al., 2023; Riesthuis, 2022), yet the study lacked sufficient statistical power to detect these differences. Another methodological limitation in eyewitness research is that the psychometric properties of laboratory measurement procedures are rarely evaluated (Lilienfeld & Strother, 2020). One explanation is that eyewitness identification studies often rely on single-item measures (i.e., one lineup), which preclude traditional reliability estimation (Nunnally, 1978), whereas the open-ended nature of free recall memory tasks complicate psychometric evaluation. In contrast, eyewitness memory studies frequently involve multiple-item tasks (e.g., 10-40 item recognition tests), for which reliability and validity can—and should—be formally examined (Lilienfeld & Strother, 2020). Assessing these properties is essential to ensure that conclusions about eyewitness memory performance, and about factors that improve or impair it, are robust and replicable (Flake & Fried, 2020). The need to assess the psychometric properties of such tasks is particularly salient given the growing interest in, for example, cross-cultural comparisons of eyewitness testimony (e.g., Hope et al., 2022), which raises critical issues such as measurement non-invariance (i.e., differential item functioning; Millsap, 2011). Currently, many researchers create their own mock crime videos and corresponding memory tests, suggesting that measurement instruments are often constructed on the fly (Flake et al., 2017) which complicates research synthesis and cumulative evidence building (Elson et al., 2023). Finally, evaluating the reliability and validity of these measures is crucial because we typically aim to make inferences about eyewitness memory as a psychological construct, not merely about memory for a specific mock crime or stimulus set. Accordingly, the present study develops a two-alternative forced-choice (2AFC) recognition memory test based on the recently proposed Jena Eyewitness Research Stimuli (JERS; Kruse & Schweinberger, 2023) and evaluates its psychometric properties within an Item Response Theory (IRT) framework. To address these measurement challenges, we adopt an IRT approach, moving beyond the Classical Test Theory (CTT) approach. Whereas CTT relies on aggregate performance scores (e.g., percent correct) that conflate participant ability with test difficulty, IRT models the probability of a correct response as a function of both the latent trait (the witness’s underlying memory ability) and the specific properties of each test item (e.g., item difficulty and discrimination). This distinction is critical: two witnesses might achieve identical total scores yet possess different underlying memory abilities depending on which specific items they answered correctly. By accounting for these item-level characteristics, IRT allows for "sample-independent" estimates of it