What are the sources of individual-level differences in risk taking, and how do they depend on the domain or situation in which the decision is being made? Psychologists currently answer such questions with psychometric methods, which analyze correlations across participant responses in survey data sets. In this article, we analyze the preferences that give rise to these correlations. Our approach uses (a) large language models (LLMs) to quantify everyday risky behaviors in terms of the attributes or reasons that may describe those behaviors, and (b) decision models to map these attributes and reasons onto participant responses. We show that LLM-based decision models can explain observed correlations between behaviors in terms of the reasons different behaviors elicit and explain observed correlations between individuals in terms of the weights different individuals place on reasons, thereby providing a decision theoretic foundation for psychometric findings. Since LLMs can generate quantitative representations for nearly any naturalistic decision, they can be used to make accurate out-of-sample predictions for hundreds of everyday behaviors, predict the reasons why people may or may not want to engage in these behaviors, and interpret these reasons in terms of core psychological constructs. Our approach has important theoretical and practical implications for the study of heterogeneity in everyday behavior. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

中文翻译：

---

使用大型语言模型探索风险承担的可变性。


冒险行为中个体差异的根源是什么？它们如何取决于决策的领域或情况？心理学家目前使用心理测量方法来回答此类问题，该方法分析调查数据集中参与者反应之间的相关性。在本文中，我们分析了引起这些相关性的偏好。我们的方法使用（a）大型语言模型（LLMs ）根据可以描述这些行为的属性或原因来量化日常风险行为，以及（b）决策模型将这些属性和原因映射到参与者的响应上。我们表明LLM基于决策模型可以根据不同行为引起的原因来解释观察到的行为之间的相关性，并根据不同个体对原因的权重来解释观察到的个体之间的相关性，从而为心理测量结果提供决策理论基础。自从LLMs可以为几乎任何自然决策生成定量表示，它们可用于对数百种日常行为进行准确的样本外预测，预测人们可能或可能不想参与这些行为的原因，并以多种方式解释这些原因核心心理结构的术语。我们的方法对于日常行为异质性的研究具有重要的理论和实践意义。 （PsycInfo 数据库记录 (c) 2024 APA，保留所有权利）。