Standard: Promoting Ethics in STEM with Transparency Initiatives that Work: An Interdisciplinary Investigation

在STEM领域产生有效的科学发现，关键取决于科学家的道德实践。研究诚信正受到越来越多的关注，因为最近出现了数据造假的例子，对无法复制已发表的结果的抱怨越来越多，以及研究表明，研究人员承认从事有损科学结果有效性的可疑行为。为了促进道德实践，许多大学和研究机构鼓励研究人员在研究过程中更加透明。透明度倡议包括鼓励研究人员预先注册他们的研究——也就是说，在开始研究之前提交一份研究计划——或者提供符合道德标准的公开证明。在防止有问题的研究实践方面，这种透明度倡议的效果如何，人们知之甚少。当前的项目探索透明度政策何时起作用，何时可能适得其反，实际上鼓励不道德的行为，以及如何改进其设计以使其更有效。研究旨在促进诚信的举措是如何与人类行为相互作用的，了解导致人们在仍认为自己诚实的情况下违反规则的背景是至关重要的。当前的跨学科项目结合了心理学、行为经济学和机器学习的见解，以促进对旨在提高研究透明度的政策如何影响个人违反道德行为倾向的理解。该项目使用机器学习分析真实世界数据、调查数据和在线实验的独特组合来探索不道德科学行为的预测因素和原因。一项研究检查了一个大型临床研究公共数据库中预先注册的研究计划。选定的研究计划与预注册研究的学术期刊文章相匹配，并且注意到由于工作的科学问题而被撤回的文章。使用自然语言处理和机器学习交叉的计算方法，比较了被撤回和未被撤回的出版物的预注册。该分析的目的是确定预先注册的研究计划的特征（例如，更短或使用更模糊的语言），这些特征预示着最终的文章将在稍后被撤回。该分析确定了善意的干预（预注册）何时不能促进负责任的研究行为。另一项研究调查了STEM研究人员关于科学不端行为的原因以及旨在防止这种行为的透明度举措的看法。这项工作还得到了在线实验的补充，这些实验旨在调查在一个任务中的不当行为心理，在这个任务中，个人有动机为了个人利益而误报信息，这类似于科学家从他们的数据中梳理出重要结果的动机。这些实验调查了鼓励个人提供道德行为证据的透明度政策可能会产生意想不到的效果的条件，增加而不是减少个人违反道德标准的倾向。最后一组实验测试了直接解决先前研究中发现的不当行为的心理驱动因素的行为干预措施。总而言之，这个跨学科项目为促进STEM道德行为的透明度倡议的设计提供了信息。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。

The production of valid scientific findings in STEM fields depends crucially on the ethical practices of scientists. Research integrity is receiving increased attention because of recent instances of data-fabrication, growing complaints about failure to replicate published results, and studies showing that researchers admit to engaging in questionable practices that undermine the validity of scientific results. To promote ethical practices, many universities and research institutions encourage researchers to be more transparent in the research process. Transparency initiatives include encouraging researchers to pre-register their studies - that is, to submit a study plan prior to beginning the research - or provide public proof of compliance with ethical standards. Little is known about how well such transparency initiatives work in terms of preventing questionable research practices. The current project explores when transparency policies work, when they may backfire, actually encouraging unethical behavior, and how to improve their design to make them more effective. It is crucial to investigate how initiatives designed to promote integrity interact with human behavior and to understand the contexts that cause people to bend rules while still thinking of themselves as honest. The current interdisciplinary project combines insights from psychology, behavioral economics, and machine learning to advance an understanding of how policies designed to promote transparency in research affect individuals' propensity to commit violations of ethical conduct. This project uses a unique combination of machine learning analysis of real-world data, survey data, and online experiments to explore the predictors and causes of unethical scientific conduct. One study examines pre-registered research plans from a large public database of clinical studies. Selected research plans are matched to the academic journal articles that resulted from the pre-registered studies, and articles that are retracted - withdrawn from the journal because of scientific problems with the work - are noted. The pre-registrations of publications that were vs. were not retracted are compared using computational methods at the intersection of natural language processing and machine learning. The goal of this analysis is to identify features of a pre-registered study plan (for example, being shorter or using more vague language) that predict that the resulting article will later be retracted. This analysis identifies when a well-intentioned intervention (pre-registration) fails to promote responsible research conduct. Another study surveys STEM researchers about causes of scientific misconduct and perceptions of transparency initiatives aimed at preventing it. This work is complemented by online experiments aimed at investigating the psychology of misconduct in a task where individuals have incentives to misreport information for private gains, which is analogous to the incentives scientists experience to tease significant results out of their data. These experiments investigate the conditions under which transparency policies that encourage individuals to provide evidence of ethical conduct might have unintended effects, increasing rather than decreasing individuals' tendency to violate ethical standards. A final set of experiments tests behavioral interventions that directly address the psychological drivers of misconduct identified in the previous studies. Taken together, this interdisciplinary project informs the design of transparency initiatives that work in promoting ethical conduct in STEM.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.