Department of Psychology, University of Massachusetts, Amherst, MA 01004-7710, USA.
Psychol Rev. 2011 Jan;118(1):155-63. doi: 10.1037/a0021774.
In "Assessing the Belief Bias Effect With ROCs: It's a Response Bias Effect," Dube, Rotello, and Heit (2010) examined the form of receiver operating characteristic (ROC) curves for reasoning and the effects of belief bias on measurement indices that differ in whether they imply a curved or linear ROC function. We concluded that the ROC data are in fact curved and that analyses using statistics that assume a linear ROC are likely to produce Type I errors. Importantly, we showed that the interaction between logic and belief that has inspired much of the theoretical work on belief bias is in fact an error stemming from inappropriate reliance on a contrast (hit rate-false alarm rate) that implies linear ROCs. Dube et al. advanced a new model of belief bias, which, in light of their data, is currently the only plausible account of the effect. Klauer and Kellen (2011) disputed these conclusions, largely on the basis of speculation about the data collection method used by Dube et al. to construct the ROCs. New data and model-based analyses are presented that refute the speculations made by Klauer and Kellen. We also show that new modeling results presented by Klauer and Kellen actually support the conclusions advanced by Dube et al. Together, these data show that the methods used by Dube et al. are valid and that the belief bias effect is simply a response bias effect.
在“使用 ROC 评估信念偏差效应：它是一种反应偏差效应”一文中，Dube、Rotello 和 Heit（2010）研究了推理的受试者工作特征（ROC）曲线的形式，以及信念偏差对不同测量指标的影响，这些指标在是否暗示曲线或线性 ROC 函数方面存在差异。我们得出结论，ROC 数据实际上是曲线的，而使用假设线性 ROC 的统计分析很可能产生 I 型错误。重要的是，我们表明，激发了许多关于信念偏差的理论工作的逻辑与信念之间的相互作用实际上是一种错误，源于对暗示线性 ROC 的不适当的对比（击中率-虚报率）的依赖。Dube 等人提出了一种新的信念偏差模型，根据他们的数据，这是目前对该效应的唯一合理解释。Klauer 和 Kellen（2011）对这些结论提出了质疑，主要是基于对 Dube 等人用于构建 ROC 的数据收集方法的推测。我们提出了新的数据和基于模型的分析，驳斥了 Klauer 和 Kellen 的推测。我们还表明，Klauer 和 Kellen 提出的新建模结果实际上支持了 Dube 等人提出的结论。这些数据表明，Dube 等人使用的方法是有效的，信念偏差效应只是一种反应偏差效应。
