Video-Recordings of Eyewitness Identification in Actual Cases: The Postdictive Value of Eyewitness Behaviors

超过70%的DNA免责案例都是错误的目击证人识别。根据目击者科学家的建议，全国许多司法管辖区已经改革了他们的鉴定程序来解决这个问题。然而，即使是最好的指认程序也不能排除所有错误的识别。因此，即使采用了最佳的指认程序，警察和检察官仍然必须努力区分可靠的指认和不可靠的指认。重要的是要确定哪些变量可以帮助警察和检察官完成这项任务。后置变量是一类特别有前途的变量。后判变量是受犯罪嫌疑人在队列中是否存在影响的变量。这些包括目击者的行为，如表达的自信程度，证人做出识别决定所需的时间，努力的可见证据，口头表达，等等。本研究的目的是研究从真实世界证人的视频记录中提取的后定位变量如何在可靠和不可靠的识别决策之间进行分类。在某种程度上，后置变量证明对区分可靠的识别和不可靠的识别是有用的，这些发现也将鼓励尚未录像识别程序的司法管辖区开始这样做。事实上，只有通过录像记录整个鉴定程序，这些司法管辖区才能确保对这些变量进行完整和准确的记录，这些变量可用于区分可靠和不可靠的鉴定。本研究的目的是确定在真实世界的阵容中，什么样的组合最适合可靠的目击者识别和不可靠的目击者识别。然后，这些发现将被用于开发一种算法，供警方和检察官在未来的调查中评估目击者指认的可靠性。加州圣克拉拉县地方检察官办公室和圣何塞警察局将提供目击者完成实际警察列队的录像。圣克拉拉县的独特之处在于，它较早地采用了最佳实践目击者识别程序（对位置变量进行有效评估的要求），最近又实施了一项对所有阵容进行录像的政策。当这些视频可用时，盲目评分者（不知道被识别的人是嫌疑人还是填充者）将评估一系列已知的定位变量。这些定位分数将被回归到结果变量是否证人识别嫌疑人或已知无辜填充。嫌疑犯识别的临界质量很可能是罪犯识别，而所有填充物的识别都是错误识别的明确实例。因此，用于区分准确识别和错误识别的后置变量应该区分可疑识别和填充识别。除了检查已知后置变量（例如，置信度，决策时间，口头话语）的预测有效性外，我们还将使用广泛的编码方案来编码许多其他目击者行为，我们将检查这些额外的目击者行为是否可以提高分类性能，超过已知后置变量所达到的性能。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Over 70% of DNA exonerations have been cases of mistaken eyewitness identification. Based on recommendations from eyewitness scientists, a large number of jurisdictions across the country have reformed their identification procedures to address this problem. Nevertheless, even the best lineup procedures fail to weed out all mistaken identifications. Hence, police and prosecutors still must attempt to sort between reliable identifications and unreliable identifications even when the best lineup procedures are followed. It is important to determine what variables can help police and prosecutors with this task. Postdiction variables are a particularly promising class of variables. Postdiction variables are variables that are influenced by the presence or absence of the guilty suspect in the lineup. These include eyewitness behaviors such as expressed level of confidence, the amount of time it takes the witness to make an identification decision, visible evidence of effort, verbal utterances, among others. The purpose of the present work is to examine how well postdiction variables extracted from video recordings of real-world witnesses making identification decisions can sort between reliable and unreliable identification decisions. To the extent that postdiction variables prove useful for sorting between reliable identifications and unreliable identifications, these findings would also encourage jurisdictions that are not yet video recording identification procedures to begin doing so. Indeed, only by video recording the entirety of the identification procedure can these jurisdictions ensure a complete and accurate record of these variables that can be used to sort between reliable and unreliable identifications.The goal of this research is to determine what combination of postdictors best sort between reliable eyewitness identifications and unreliable eyewitness identifications in real-world lineups. The findings will then be leveraged to develop an algorithm that police and prosecutors can use to assess the reliability of eyewitness identifications in future investigations. The District Attorney’s Office of Santa Clara County, CA and the San Jose Police Department will provide video-recordings of witnesses completing actual police lineups. Santa Clara County is unique as an early adopter of best-practice eyewitness identification procedures (a requirement for valid assessment of postdiction variables) and more recently implementing a policy of video-recording all lineups. As these videos become available, blind scorers (blind as to whether the identified person is the suspect or a filler) will assess a host of known postdiction variables. These postdiction scores will then be regressed on the outcome variable of whether the witness identified the suspect or a known-innocent filler. A critical mass of suspect identifications are likely to be culprit identifications whereas all identifications of fillers are definitive instances of mistaken identifications. Accordingly, postdiction variables that are useful for separating accurate from mistaken identifications should distinguish between suspect identifications and filler identifications. In addition to examining the predictive validity of known postdiction variables (e.g., confidence, decision time, verbal utterances), we will use an extensive coding scheme to code for numerous other eyewitness behaviors and we will examine whether these additional eyewitness behaviors can improve classification performance over and above the performance achieved with know postdiction variables.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.