Signal Detection Theory in Single-Trial Human-Decision Tasks

关于检测理论的文献几乎完全集中在信号和噪声试验混合的重复测量任务上。对重复测量任务的依赖在很大程度上是务实的(例如，增加统计能力)。但是，现实世界中有一些重要的单项试验任务没有得到适当的探索，往往表现不同。我最近发现了两个与此类任务的诊断价值有关的发现。首先，填充物控制方法，即用已知的噪声刺激包围感兴趣的刺激，不会提高辨别能力，但可以提高诊断价值。因为对已知噪声刺激的肯定是可检测的误差，所以它们不像临界噪声误差那样重要。此外，由于已知噪声刺激比真正的阳性信号更能减少临界噪声误差，因此它们提高了诊断价值。其次，遇到纯噪音测试的人往往会使用比遇到信号测试的人更低的标准。既然已经知道了这种差异，人们就可以设计出减少差异并提高诊断价值的干预措施。*测试和开发这些发现的具体背景是目击者记忆，但修订后的检测模型表明，这些一般原则适用于其他单项试验任务。该模型表明，填充物控制方法可以使许多检测任务，如取证对象识别和取证检验等受益。此外，这些任务可能受到信号/噪声标准差异的困扰，并可能从减少差异中受益。通过计算模型和实验，本研究将完善对目击者情境的理解，并将这些发现扩展到其他单项试验任务。在目击者的背景下，我将研究这些现象是如何通过记忆强度、填充物的数量、说明和呈现方法来调节的。此外，我将通过将该模型应用于其他应用环境(例如，指纹检查、笔迹分析、咬痕分析)来演示这些发现的一般性。*这项研究将引起科学界和法律界的兴趣。对加拿大人来说，好处是一个运行更好的法律体系，以及一种潜在的高级诊断测试，可以在一系列不同的应用环境中使用。主要成果是更好地理解人类认知和探测理论。事实上，如果不将检测理论应用于单次试验任务，就不会发现信号/噪声标准的差异。这种差异很可能不存在于重复测量的任务中。

The literature on Detection Theory has almost exclusively focused on repeated-measures tasks in which signal and noise trials are intermixed. The reliance on repeated-measures tasks is largely pragmatic (e.g., increased statistical power). But, the real world has some important single-trial tasks that have not been properly explored and tend to behave differently. I have recently made two discoveries related to diagnostic value in such tasks. First, the filler-control method, which involves surrounding a stimulus of interest with known-noise stimuli, does not improve discriminability but can improve diagnostic value. Because affirmatives on known-noise stimuli are detectable errors they are not as consequential as critical-noise errors. Further, because known-noise stimuli decrease critical-noise errors more than true positives, they improve diagnostic value. Second, a person who encounters a noise-only trial will tend to use a lower criterion than will someone who encounters a signal trial. Now that this discrepancy is known, one can devise interventions that reduce discrepancy and improve diagnostic value. ******The specific context in which these discoveries have been tested and developed is eyewitness memory, but the revised detection model shows that these general principles are applicable to other single-trial tasks. The model shows that many detection tasks, such as forensic-object recognition and forensic examination could benefit from the filler-control method. Further, these tasks are likely plagued by signal/noise criteria discrepancy and could benefit from reduced discrepancy. Through computational modeling and experimentation, the present research will refine understanding in the eyewitness context and extend these findings to other single-trial tasks. Within the eyewitness context, I will examine how these phenomena are moderated by memory strength, number of fillers, instructions, and presentation methods. Further, I will demonstrate the generality of these discoveries by applying this model to other applied contexts (e.g., fingerprint examination, handwriting analysis, bite-mark analysis). ******The research will be of interest to the scientific and legal communities. The benefit to Canadians is a better functioning legal system and potentially a superior diagnostic test that can be used in an array of different applied settings. The major deliverable is a better understanding of human cognition and Detection Theory. Indeed, without applying Detection Theory to single-trial tasks, signal/noise criteria discrepancy would not have been discovered. This discrepancy likely does not exist in repeated-measures tasks.