Assessing Eyewitness Identification

该项目将建立一个数据库，总结过去50年发表的1000多项面部识别研究的结果。利用这个数据库，本项目将解决四个研究问题：1.总结80个因素对人脸识别性能的影响。2.将通过测试研究方法的差异是否影响研究结果的模式来检验目击者实验室研究结果对实际目击者表现的概括性(应用)。3.将评估新的证人表现衡量标准(基于信号检测理论)的影响，以确定与以往的表现衡量标准相比，这些衡量标准是否能充分说明研究结果。4.研究不同研究人员提出的60多个目击证人绩效指标之间的关系模式。预计这些措施是高度多余的，目击者的表现可以在少数几项易于理解的措施中得到充分反映。该项目将建立一个全面的面部识别研究数据库，其中包括过去50年发表的1000多项研究；它将扩展彭罗德和他的同事在1986年和2002年建立的数据库。利用这个数据库，将解决四个研究问题，这四个问题对于更好地了解目击者的表现至关重要。这些问题如下：首先，研究人脸识别研究中某些自变量对识别性能的影响。累积的数据库包括大约80个独立变量，这些变量便于审查每个变量对识别绩效的相对贡献。这不仅有助于总结以前关于面部识别的文献的发现，而且通过揭示值得特别关注的有价值的变量，为未来的研究指明了方向。其次，将检验目击者实验室研究结果对真实世界中目击者实际表现的概括性。一些研究人员警告说，由于不同研究方法的同质性和研究方法的外部有效性较低，目击者研究结果的概括性存在风险。为了测试结果的概括性，本项目将调查研究方法(例如，参与者类型、刺激类型、实验程序等)缓和独立变量对目击者表现的影响，如果是这样的话，实验室研究是否低估或高估了真实世界的表现。第三，将测试一种新的信号检测模型对证人在列队期间的表现的影响，并通过对累积数据库的分析来测试其有效性。过去基于信号检测理论的目击者研究通常集中在识别有罪嫌疑人(即命中率)和无辜嫌疑人(即虚警率)，而忽略了填充者身份(这会带来代价，如“烧死”证人、失踪的肇事者和未能清除无辜嫌疑人)。与以往的研究相比，提出的信号检测模型将阐明填充物识别在目击证人行为中的理论和实践重要性。最后，本研究将探讨以累积资料库衡量目击表现指标的心理测量学结构。虽然研究人员经常使用极少的选定指标(如DR、PPV等)来衡量目击者的表现，但有60多个指标衡量或与目击者的表现有关。预计目击者的表现反映在少数几个组成部分中，每个指数在不同程度上衡量一个或多个组成部分。该项目旨在揭示各种指数是如何相互联系的，也就是以可获得性的术语揭示这些指数衡量的目击者表现的哪些组成部分，并就哪些(最好是可获得性的)衡量标准最好地表征绩效的哪些方面得出建议。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project will build a database summarizing the results of over 1,000 facial identification studies published in the last five decades. Using this database, the project will address four research questions: 1. The effect of 80 factors on facial identification performance will be summarized. 2. The generalizability (application) of findings from eyewitness laboratory research to actual eyewitness performance will be examined by testing whether variations in research methods affect patterns of research results. 3. The implications of new witness performance measures (based on signal detection theory) will be assessed to see if they provide an adequate account of research findings as compared to prior performance measures. 4. The pattern of relationships among more than 60 indices of eyewitness performance proposed by various researchers will be investigated. It is anticipated that these measures are highly redundant and eyewitness performance can be fully reflected in a handful of readily understood measures. This project will build a comprehensive database of facial identification research that includes over 1,000 studies published in the last five decades; it will expand the database constructed by Penrod and his colleagues in 1986 and 2002. Using this database, four research questions, which are crucial for better understanding eyewitness performance, will be addressed. These questions are as follows: First, the effect of certain independent variables in facial identification research that influence identification performance will be investigated. The accumulated database includes approximately 80 independent variables that facilitate a review of the relative contribution of each variable to identification performance. This will not only help summarize findings from previous literature on facial identification, but also suggest a direction for future research by revealing valuable variables deserving particular attention. Second, the generalizability of findings from eyewitness laboratory research to actual eyewitness performance in the real world will be tested. Some researchers warn of the risk to the generalizability of eyewitness research results because of methodological homogeneity across studies and low external validity in research methodology. To test the generalizability of finings, this project will investigate whether research methodology (e.g., participant type, stimulus type, experimental procedures, etc.) moderates the effects of independent variables on eyewitness performance and if so, whether laboratory research under-estimates or over-estimates real-world performance. Third, the implications of a new signal detection model for eyewitness performance during lineups and test its validity through analysis of the cumulative database will be tested. Past eyewitness research based on signal detection theory has generally focused on identification of a guilty suspect (i.e., hit rate) and an innocent suspect (i.e., false alarm rate) while ignoring filler identifications (which carry costs such as "burning" witnesses, missing perpetrators, and failing to clear innocent suspects). In contrast to past research, the proposed signal detection model will clarify the theoretical and practical importance of filler identifications in eyewitness performance. Finally, the psychometric structure of indices measuring eyewitness performance with the cumulative database will be investigated. Although researchers have frequently used very few selected indices (e.g., DR, PPV and others) to measure eyewitness performance, there are more than 60 indices measuring or relevant to eyewitness performance. It is anticipated that eyewitness performance is reflected in a handful of components and that each index measures one or multiple components to a different degree. The project aims to reveal how the various indices relate to one another, that is, reveal in accessible terms--what components of eyewitness performance these indices measure, and arrive at suggestions about which (preferably accessible) metrics best characterize which aspects of performance.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.