Collaborative Research: Using Causal Explanations and Computation to Understand Misplaced Beliefs

因果解释为事件发生的原因提供了答案。对因果解释的信念（例如，投资新技术将使你的退休可用资金增加）指导未来的行为（例如，投资加密货币）。但人们可以——而且经常——相信对世界的因果解释是错误的。理解错误的因果解释的本质是什么使其可信，这对于教导人们拒绝对事件的错误解释至关重要。在这项工作中，pi调查了是什么让事件的错误因果解释吸引人们，是什么鼓励人们接受这些错误的信念。持有不正确的因果解释可能会产生严重的破坏性影响，例如人们寻求无效的健康治疗或投资于没有回报的财务策略。因此，重要的是要更好地理解是什么使错误的因果解释具有吸引力，以便可以部署策略来抵消它们的采用。在这项工作中，pi进行了一系列研究，以提供对因果解释的本质使其具有吸引力的深刻理解。在研究中，pi探索了许多不同的解释因果因素。使用他们的结果，pi然后做一个初步的尝试，以减少对不正确的因果解释的认可。这项研究通过让学生参与具有强大转化成分的研究，对科学产生了更广泛的影响。这样的研究可以帮助学生将科学与现实世界联系起来，培养他们对科学和批判性思维的兴趣。此外，拟议的工作将对科学素养产生更广泛的影响，因为它隔离了科学解释中可能使它们更或更不可能被相信的东西。pi使用因果解释文献中的心理学方法来研究对各种错误和不正确的因果解释的看法。这些方法包括让人们阅读解释，并对解释的说服力、满意度和可信度进行评分。此外，在这些研究中，参与者对解释的因果结构做出判断，例如解释包含多少因果因素，解释有多复杂，解释可以解释多少事件。pi使用大量在线参与者的样本，以确保有许多不同信仰的人被包括在研究中。在每项研究中，pi让参与者对不正确的因果解释（例如，“吃糖是2型糖尿病的主要原因”）以及基于事实的同一事件的因果解释（例如，“2型糖尿病有多种原因，包括超重和遗传易感性”）进行评分。这种比较可以分离出错误解释的独特之处。使用机器学习，研究人员调查了事实和错误信息解释的特征结构的程度，超出了它们的感知方式（例如，因果结构的复杂性），这可能允许在这些解释类型之间进行更多的自动区分。最后，pi利用他们的发现创建了一套行为研究，在这些研究中，他们改变了解释的呈现方式，以探索呈现对解释认可的影响程度。具体来说，pi创造了新的因果解释，操纵了在早期实验中最能预测错误因果信念的因果因素（例如，复杂性，因果因素的数量）。我们的目标是看看是否通过改变这些重要的因果因素，可以减少对错误信念的认可。通过这些研究，我们可以更广泛地了解如何防止吸收不正确的信息，以支持基于事实的解释。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Causal explanations provide answers to why an event happened. Beliefs in causal explanations (for example, investing in new technologies will cause your available money for retirement to increase) guide which behaviors to engage in for the future (e.g., invest in crypto-currency). But people can – and often do – believe causal explanations of the world that are wrong. Understanding what in the nature of an incorrect causal explanation makes it believable is critically important for teaching people to reject incorrect explanations of events. In this work, the PIs investigate what makes incorrect causal explanations of events appealing to people and what encourages the adoption of these misplaced beliefs. Holding incorrect causal explanations can have critically damaging effects, such as people pursuing health treatments that are ineffective or investing in financial strategies that do not pay out. It is therefore important to better understand what makes incorrect causal explanations appealing so strategies can be deployed to counteract their adoption. In the work, the PIs conduct a series of studies to provide a strong understanding of what in the nature of a causal explanation makes it appealing. Across studies, the PIs explore many different causal elements of explanations. Using their results, the PIs then make a preliminary attempt to reduce endorsement of incorrect causal explanations. This research has a broader impact on science by involving students in research that has a strong translational component. Such research helps students connect science to the real-world, growing their interest in science and critical thinking at large. Additionally, the proposed work will have broader impacts on science literacy by isolating what in scientific explanations may make them more or less likely to be believed. The PIs use psychological methods from the causal explanation literature to study perceptions of a wide range of misplaced and incorrect causal explanations. These methods include having people read explanations and rate how compelling, satisfying, and believable the explanations are. In addition, participants in these studies make judgments about the causal structure of the explanations, such as how many causal factors the explanations include, how complex the explanations are, and how many events the explanations can explain. The PIs use large samples of online participants to ensure that people with many different beliefs are being included in the studies. In each study, the PIs have participants rate incorrect causal explanations (e.g., “eating sugar is the main cause of type 2 diabetes”) as well as fact-based causal explanations of the same events (e.g., “type 2 diabetes has multiple causes, including being overweight and having a genetic predisposition”). This comparison allows for isolation of what is unique about misplaced explanations. Using machine learning, the researchers investigate the degree to which there are characteristic structures of factual and misinformation explanations, beyond how they are perceived (e.g., the complexity of causal structure), which may allow for more automatic differentiation between these explanation types. Finally, the PIs use their findings to create a set of behavioral studies where they alter how explanations are presented to explore the degree to which presentation impacts endorsement of the explanations. Specifically, the PIs create new causal explanations that manipulate the causal elements that were most predictive of endorsement for incorrect causal beliefs in earlier experiments (e.g., complexity, number of causal factors). The goal is to see if by changing these important causal elements, endorsement of incorrect beliefs can be reduced. Through these studies we can learn more generally how to prevent the uptake of incorrect information in favor of fact-based explanations.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.