Analyzing Student Usage and Outcome Data from a Randomized Experiment of Cognitive Tutor Algebra

代数涉及的数学对STEM概念的进一步研究和许多技术职业的工作至关重要。然而，许多高中生很难通过代数考试。认知代数辅导1 （CTA1）是一门课程，包括教科书和一个自动化的计算机应用程序，旨在向学生提供个性化的指导。该项目将建立在一项随机对照实验的基础上，该实验检验了CTA1的有效性。研究人员将通过检查学生使用该程序时的行为和进步日志模式、教师调查数据和学生成绩数据，来研究CTA1实现其效果的机制。该项目将采用贝叶斯统计模型来推断是否存在因果模型来解释积极的学习成果和课程使用之间的关系。pi将使用一种特殊形式的中介分析，即鲁宾因果模型，来发展和检验关于学生在有CTA1和没有CTA1时会经历什么情况的假设。这种分析方法支持在处理分配（CTA1）和结果（学生成绩）之间关系的中介变量的识别。该研究还测试了一种特殊形式的中介分析，即主要分层，以确定其在进行这种分析中的效用。多种中介模型将被检查，以确定教育研究人员如何可能开发更好的实验研究。

Algebra addresses mathematics that is crucial for further study of STEM concepts and for work in many technical careers. Yet many high school students struggle to pass Algebra. Cognitive Tutor Algebra 1 (CTA1) is a curriculum that includes both textbook components and an automated computer application that is designed to deliver individualized instructions to students. This project will build on the findings of a randomized control experiment that examined the effectiveness of CTA1. The researchers will study the mechanisms by which CTA1 achieves its effect by examining patterns in logs of students' actions and progress while they use the program, teacher survey data and student achievement data. The project will employ Bayesian statistical models to infer if causal models exist that explain the relationship between positive learning outcomes and the use of the curriculum.The PIs will use a specific form of mediation analysis, the Rubin Causal Model, to develop and test hypotheses about what the students would have experienced with CTA1 versus what they would have experienced in its absence. This analysis methodology supports the identification of mediating variables on the relationship between a treatment assignment (CTA1) and an outcome (student achievement). The study also tests the use of a particular form of mediation analysis, principal stratification, to determine its utility in conducting such analyses. Multiple mediation models will be examined to determine how educational researchers might develop better experimental research studies.