Statistical Methods for Response Process Data

技术的发展使收集各种数据成为可能，但也对统计分析提出了挑战。本研究计划旨在发展方法，分析最近以电脑为基础的教育评估所产生的回应过程资料。这些数据提供了关于考生行为的详细信息，而传统的项目反应数据无法捕捉这些信息。然而，数据格式的复杂性和人类行为的多样性给系统有效地利用这些信息带来了挑战。该项目将开发创新的方法来理解和识别学习和解决问题的个体差异。所得的信息将有助于设计个性化的指导或干预策略，以支持学生的成功，并倡导教育的包容性和公平性。此外，该项目还将开发供从业人员使用的用户友好软件，并为研究生和本科生提供研究培训机会。响应过程数据是一种新兴的数据类型，用于跟踪应答者与基于计算机的项目的交互。该项目旨在提供创新的、可扩展的、可解释的统计方法，以利用响应过程数据中的丰富信息。具体而言，该项目将重点开发1)数据驱动方法，用于从过程数据中提取特征；2)潜在变量模型，用于理解响应过程动态如何由应答者的潜在特征驱动；3)过程上的标量回归模型，用于描述响应过程与其他观察变量之间的统计关系。新的计算算法将被设计用于统计推断。模型的强可解释性将打开以前基于机器学习的过程数据方法所创建的黑箱，使验证结果更容易，并更深入地了解学生的问题解决行为。该项目的成果将使教育工作者能够更好地评估学生并设计有效的教育策略。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 development of technology allows for the collection of diverse data but also poses challenges in statistical analysis. This research project aims to develop methods for analyzing response process data generated from recent computer-based educational assessments. Such data provide detailed information on test-takers behaviors that traditional item response data cannot capture. However, the complex format of the data and the diversity of human behaviors make it challenging to utilize the information systematically and efficiently. This project will develop innovative methods to understand and identify individual differences in learning and problem-solving. The resulting information will be valuable for designing individualized instruction or intervention strategies to support student success and advocate inclusiveness and equity in education. Additionally, user-friendly software will be developed for practitioners' use, and this project will provide research training opportunities for graduate and undergraduate students.Response process data are an emerging type of data that tracks a respondent's interaction with computer-based items. This project aims to provide innovative, scalable, and interpretable statistical methods for utilizing rich information in response process data. Specifically, this project will focus on developing 1) a data-driven method for extracting features from process data, 2) a latent variable model for understanding how response process dynamics are driven by respondents' latent traits, and 3) a scalar-on-process regression model for describing statistical relationships between response process and other observed variables. Novel computational algorithms will be designed for statistical inference. The strong interpretability of the models will open the black box created by previous machine-learning-based approaches for process data, making it easier to validate the results and gain a deeper understanding of students' problem-solving behaviors. The outcomes of this project will enable educators to better evaluate students and design effective educational strategies.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.