Do people use their statistical expertise selectively to reach preferred conclusions when evaluating scientific evidence, with those more expert showing more preferential bias? We investigated this motivated numeracy account of evidence evaluation but came to a different account for biased evaluation. Across three studies (N = 2,799), participants interpreted numerical data from gun control intervention studies. In Studies 1 and 2, participants reached accurate conclusions more frequently from scientific data when those data aligned with their political preferences than when they did not, an attitude congeniality effect. This bias was unrelated to numerical ability (i.e., numeracy) and cognitive effort, although each variable predicted correct reasoning independently. Probing further, we found that attitude congeniality did not prompt people to discover valid statistical rationales for their more frequent correct conclusions. Rather, people came to right conclusions more often but for wrong reasons, suggesting why numerical ability need not be related to the congeniality effect. In Study 2, we showed this pattern was not due to forced guessing. In Study 3, we showed that the rationales, whether right or wrong, carried some weight over multiple scenarios, indicating that participants were not just expressive responding-that is, simply stating preferred conclusion regardless of the data. Statistical training did not reduce attitude congeniality biases. We suggest that people engage in "expressive rationalization" rather than "rationality" to reach preferred conclusions, finding convenient rationales for preferred conclusions that need not be valid, even though they can lead to conclusions that are. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

中文翻译：

---

对数值科学证据的偏见解释背后的心理机制。


人们在评估科学证据时是否有选择地使用他们的统计专业知识来得出首选结论，而那些更专业的人表现出更多的优先偏见？我们调查了这种证据评估的动机计算方法，但得出了不同的偏倚评估方法。在三项研究 （N = 2,799） 中，参与者解释了枪支管制干预研究的数值数据。在研究 1 和 2 中，当数据与他们的政治偏好一致时，参与者更频繁地从科学数据中得出准确的结论，这是一种态度亲和效应。这种偏倚与数字能力（即计算能力）和认知努力无关，尽管每个变量都独立地预测了正确的推理。进一步探究，我们发现态度亲和性并不能促使人们为他们更频繁的正确结论发现有效的统计理由。相反，人们更经常得出正确的结论，但出于错误的原因，这表明为什么数字能力不需要与亲和效应相关。在研究 2 中，我们表明这种模式不是由于强迫猜测造成的。在研究 3 中，我们表明，无论是对还是错，基本原理在多个场景中都有一定的分量，这表明参与者不仅仅是表达性的回应——也就是说，无论数据如何，都简单地陈述首选结论。统计训练并没有减少态度亲和性偏差。我们建议人们进行 “表达性合理化 ”而不是 “理性 ”来得出首选结论，为首选结论寻找方便的理由，这些结论不一定是有效的，即使它们可以导致有效的结论。（PsycInfo 数据库记录 （c） 2024 APA，保留所有权利）。