Treatment Effect Estimation in Cluster Randomized Experiments in the Presence of Partial Implementation

Guanglei Hong

This study will develop methods for causal inference for cluster randomized experiments in which educational innovations are implemented with varying levels of fidelity among experimental schools. In the presence of partial implementation, the intent-to-treat effect estimate—typically computed as the mean difference in student outcome between experimental schools and control schools—underestimates the effect of a fully implemented treatment. The project will focus on estimating (1) the effect of an innovation on students’ learning outcomes when the innovation is implemented with high quality, and (2) the effect when the innovation is partially implemented. I will contrast two different statistical perspectives on the implementation problem, and will adapt these analytic approaches to the multi-level data from the national randomized evaluation of Success for All, an influential comprehensive school reform program. I expect that methodological developments and empirical results from this study will inform program evaluation and program improvement.

About Guanglei Hong

N/A