全球智库信息集成服务系统

In some tests of educational interventions, individual students are randomized directly to the treatment or control group, and both intervention and control protocols are administered in an individual setting. Such an experiment is an Individual-Level Randomized Controlled Trial (I-RCT). In other tests, clusters of students (e.g., classrooms) are randomized. This sort of experiment is called a Cluster Randomized Controlled Trial (C-RCT). However, in some designs, students in the treatment group are clustered like those in a C-RCT, but students in the control group are unclustered, like students in an I-RCT. This design is called a Partially Nested Randomized Controlled Trial (PN-RCT). It is partially nested because students in the treatment group are nested in some higher level unit, such as a tutoring group or class, but students in the control group are not nested as part of the experimental design.

This paper, commissioned by the National Center for Education Research, provides readers with an introduction to PN-RCTs and ways to design and analyze the results from them. This paper was written primarily for applied education researchers with introductory knowledge of quantitative impact evaluation methods. However, those with more advanced knowledge will also benefit from some of the technical examples and appendices.

• Chapters 1 and 2 define PN-RCTs and address design issues such as possibilities for random assignment, cluster formation, statistical power, and confounding factors that may mask the contribution of the intervention.
• Chapter 3 discusses basic statistical models that adjust for the clustering of treatment students within intervention clusters, associated computer code for estimation, and a step-by-step guide, using examples, on how to estimate the models and interpret the output.
• Chapter 4 and the technical appendixes discuss more advanced statistical topics pertaining to PN-RCTs.