An Explanatory Machine Learning Framework for Teacher Effectiveness in STEM Education

本项目旨在通过开发解释性机器学习方法来研究STEM教育中的教学有效性，从而为国家利益服务。有一致的证据表明，教师的有效性差异很大，但传统的分析方法在很大程度上未能解释教学和教师差异的原因以及在什么情况下差异。解释性机器学习方法在发展公平有效的STEM教学的基础知识和理论方面具有巨大潜力，因为它们可以追踪教学理论固有和隐含的复杂特征、过程和模式，而传统方法无法做到这一点。在这个项目中，我们研究了我们可以在多大程度上利用机器学习方法来识别和解释产生学生学习的概况，途径和实践（例如，教师是谁，教师知道什么，教师相信什么，感知和体验什么，教师做什么），以及这些概况和实践在STEM教育背景下如何变化。该项目的成果有可能加速对有效教学、教师准备、教师发展和学生学习的理论和实践的研究。这是一个为期三年的BCSER: STEM学习和学习环境研究中的STEM教育研究个人研究者发展项目。STEM教育和教师发展领域在发展复杂的教与学理论以及支持和实施这些理论研究的工具和措施（例如，教师知识、文化响应型教师自我效能感、课堂观察）方面取得了实质性进展。然而，最近的文献指出，我们在有效教师和教学理论中发现的复杂性与我们用来分析这些理论的流行方法之间存在不匹配。例如，理论认为教学是一种高度互动、自适应、非线性和情境依赖的实践；然而，该领域几乎完全利用了简单的线性回归模型，这些模型无法轻易地检测和分析这些复杂的模式。越来越多的人认识到需要制定、开发和发展针对STEM教学和学习研究目的的方法。本项目旨在通过开发和调整解释性机器学习方法（例如神经网络）来填补这一空白，以分析教学有效性的研究，并检查这些方法在多大程度上可以预测、解释和以优于传统方法的方式将有效教学情境化。研究结果有可能通过识别复杂的特征和背景内和跨背景的有效教学概况，以及开发广泛适用于STEM教育研究的可扩展机器学习方法，广泛地建立能力并影响该领域。该项目得到了比尔和梅林达·盖茨基金会、施密特期货和沃尔顿家族基金会的合作支持。该项目还得到了美国国家科学基金会EDU STEM教育研究核心研究能力建设（ECR: BCSER）项目的支持，该项目旨在培养研究人员在STEM学习和学习环境的核心领域开展高质量STEM教育研究的能力，扩大STEM领域的参与，以及STEM劳动力的发展。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。

This project aims to serve the national interest by developing explanatory machine learning methods for the study of teaching effectiveness in STEM education. There is consistent evidence that teachers vary widely in their effectiveness but conventional analytic methods have largely failed to explain why and under what contexts teaching and teachers vary. Explanatory machine learning methods hold significant potential in developing fundamental knowledge and theories of equitable and effective STEM teaching because they can track complex features, processes and patterns inherent in and implied by theories of teaching in ways where conventional methods fall short. In this project, we examine the extent to which we can leverage machine learning methods to identify and explain profiles, pathways and practices (e.g., who teachers are, what teachers know, what teachers believe, perceive and experience, what teachers do) that produce student learning and how these profiles and practices vary across STEM education contexts. The outcomes of this project have the potential to accelerate research on the theory and practice of effective teaching, teacher preparation, teacher development and student learning. This is a three-year BCSER: Individual Investigator Development project in STEM Education Research within Research on STEM Learning and Learning Environments.The fields of STEM education and teacher development have made substantial progress in developing sophisticated theories of teaching and learning and instruments and measures that support and operationalize research on those theories (e.g., teacher knowledge, culturally responsive teacher self-efficacy, classroom observations). Recent literature has, however, noted that there is a mismatch between the complexity found in our theories of effective teachers and teaching and the prevailing methods we use to analyze those theories. For example, theories suggest that teaching is a highly interactive, adaptive, nonlinear and context-dependent practice; yet the field has almost exclusively drawn on simple linear regression models that cannot readily detect and analyze these complex patterns. There is a growing recognition of the need to craft, develop and grow methodologies specific to the purposes of STEM teaching and learning research. This project aims to fill this gap by developing and adapting explanatory machine learning methods (e.g., neural networks) to analyze studies of teaching effectiveness and examining the extent to which these methods can predict, explain and contextualize effective teaching in ways that outperform conventional methods. The results have the potential to broadly build capacity and impact the field by identifying complex features and profiles of effective teaching within and across contexts and developing scalable machine learning methods that are broadly applicable to STEM education studies. This project is supported through a partnership with the Bill & Melinda Gates Foundation, Schmidt Futures, and the Walton Family Foundation. This project is also supported by NSF's EDU Core Research Building Capacity in STEM Education Research (ECR: BCSER) program, which is designed to build investigators' capacity to carry out high-quality STEM education research in the core areas of STEM learning and learning environments, broadening participation in STEM fields, and STEM workforce development.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.