The wisdom-of-crowds effect describes how aggregating judgments of multiple individuals can lead to a more accurate judgment than that of the typical—or even best—individual. However, what if there are no other individuals’ judgments at one’s disposal? We investigated when an individual can avail theirself of the wisdom of their “inner crowd” to improve the quality of their confidence judgments by either (a) averaging their two confidence judgments or (b) selecting the higher of the two (i.e., maximizing) in two-alternative choice tasks. In a simulation analysis based on a signal detection model of confidence, we investigated how the “kindness” versus “wickedness” of items (i.e., the degree to which the majority of people chooses the correct or wrong answer) affect the performance of averaging and maximizing. Simulation and analytical results show that irrespective of the type of item, averaging consistently improves confidence judgments, but maximizing is risky: It outperformed averaging only once items were answered correctly 60% of the time or more—a result, which has not been established in prior work. We investigated the relevance of these effects in three empirical datasets since a person’s actual confidence judgments are redundant (median correlations ranged between .5 and .85). Averaging two confidence judgments from the same person was superior to maximizing, with Cohen’s d’s effect sizes ranging from 0.67–1.44. As people typically have no insight about the wickedness of the individual item, our results suggest that averaging—due to its robustness—should be the default strategy to harness one’s conflicting confidence judgments.

中文翻译：

---

“内部群众的智慧”如何提高信心判断的准确性。

群体智慧效应描述了多个人的综合判断如何导致比典型的甚至最好的个人的判断更准确的判断。但是，如果没有其他人的判断可供自己支配呢？我们调查了个人何时可以通过 (a) 平均他们的两个信心判断或 (b) 选择两者中的较高者（即最大化）来利用他们“内部人群”的智慧来提高他们的信心判断的质量在二选一的选择任务中。在基于置信度信号检测模型的模拟分析中，我们调查了项目的“善意”与“恶行”（即大多数人选择正确或错误答案的程度）如何影响平均和最大化。模拟和分析结果表明，无论项目的类型如何，求平均值都始终如一地提高了置信度判断，但最大化是有风险的：只有在 60% 或更多的时间正确回答项目时，它的表现才优于平均——这个结果在之前的工作。我们在三个经验数据集中调查了这些影响的相关性，因为一个人的实际信心判断是多余的（中值相关性介于 0.5 和 0.85 之间）。对来自同一个人的两次置信判断进行平均优于最大化，Cohen 的 d 效应大小范围为 0.67–1.44。由于人们通常不了解单个项目的邪恶程度，因此我们的结果表明，平均 - 由于其稳健性 - 应该是利用一个人相互矛盾的信心判断的默认策略。