The Variable Response Model: Theory, Implementation, and Application

对心理能力的评估在日常生活中无处不在，包括工作面试、驾驶考试和大学入学考试，仅举几例。心理能力可以借助统计模型和测试中可观察到的行为来估计。通常，这些模型针对特定的响应格式进行定制，例如多项选择、真假或连续响应。相反，本项目引入了可变响应模型，灵活地考虑了同一测试中的各种响应格式。与许多类似的模型相比，可变反应模型进一步解释了测试问题的正确答案未知的情况。例如，这适用于证人的证词。当几名犯罪证人接受采访时，正确的回答代表了犯罪的真实细节。可变反应模型提供了对这些细节的估计，同时考虑了证人报告犯罪能力的个体差异。在目前的研究项目中，我首先关注理论，旨在使可变反应模型适应多种心理能力导致相同测试成绩的情况。该项目的第二个目标是在软件中实现模型，允许从业者在尽可能少的训练或先验知识的情况下应用模型。最后，将变量响应模型应用于证人证言。具体来说，我的目标是通过考虑证人描述犯罪者能力的个体差异来提高面部合成的质量。最终，该项目可能有助于建立可变反应模型，使其成为同一罪行有多个证人在场时的标准法医工具。可变响应模型还可以帮助教育测试，其中混合响应格式为从业者提供了更好的灵活性。我进一步期望目前的项目能引起人类学的兴趣，在人类学中，类似但可能不那么强大的方法已被广泛使用。

The assessment of mental abilities is ubiquitous in everyday life and includes job interviews, driving exams, and university admission tests, to name just a few. Mental abilities can be estimated with the help of statistical models and the observable behavior in a test. Typically, these models are tailored to specific response formats such as multiple choice, true-false, or continuous responses. Conversely, the present project introduces the Variable Response model which flexibly accounts for various response formats within the same test. In contrast to many similar models, the Variable-Response model further accounts for situations where the correct responses to the test questions are unknown. This applies to witness testimony, for example. When several witnesses of a crime are interviewed, the correct responses represent the true details of the crime. The Variable Response model provides estimates of these details while accounting for inter-individual differences in the witnesses´ abilities to report about the crime. In the present research project, I first focus on theory and aim to adapt the Variable Response model to situations where multiple mental abilities contribute to the same test scores. The second goal of the project is to implement the model in software that allows practitioners to apply the model with as little training or prior knowledge as possible. Finally, the project applies the Variable Response model to witness testimony. Specifically, I aim to improve the quality of face composites by accounting for inter-individual differences in the witnesses´ ability to describe the perpetrator. Ultimately, the project may help to establish the Variable Response model as a standard forensic tool when several witnesses of the same crime are available. The Variable Response model could also aid educational testing where the mixing of response formats provides improved flexibility to practitioners. I further expect the present project to be of interest in Anthropology where similar but arguably less powerful methods have been used extensively.