Science Outcome Design Parameters for Cluster-Randomized Trials Involving Teachers

Abstract

Programs aiming to improve student science outcomes often involve teachers. When designing cluster-randomized trials (CRTs) to evaluate the causal effects of these programs, investigators need empirical design parameters as critical inputs in the study design phase. Using proper empirical design parameters to conduct power analysis, investigators can ensure designs have adequate statistical power to detect treatment effects and efficiently use resources. The current study uses two nationally representative samples to estimate empirical design parameters for science outcomes in settings of students (nested within teachers) nested within schools. We compile design parameters for intraclass correlation coefficients and proportions of variance explained by covariates. We illustrate how to use the results to design efficient and effective CRTs in both the conventional statistical power analysis framework and the optimal design framework that additionally considers sampling costs.