Eric Hedberg

Education policies and interventions are often implemented in ways that do not render it feasible to use a randomized control trial to test their effects. The purposes of this study are to derive theory for computing exact statistical power for three common quasi-experimental designs (QEDs) - nonequivalent control group designs, difference-in-differences, and interrupted time series - and to develop software for computing statistical power using the derived approaches for those QEDs.

In a multilevel model, the statistical power, precision of estimates of treatment effects, the most efficient allocation of sample between levels, and the minimum detectable effect size all depend on the intraclass correlation structure and the effectiveness of any covariates in explaining variation at each level where they are used.

Co-Principal Investigator: Eric Hedberg (NORC) In many states, the data from statewide longitudinal data systems (SLDS) are available to a small group of state education agencies and scholars.

This project used five state data systems to compute design parameters (including intraclass correlations and variance accounted for by pretest data) at the state, district, school, and classroom levels. Additionally, meaningful subsets of the states defined geographically, in terms of achievement levels (e.g., low achieving schools or districts), and in terms of socioeconomic status (e.g., low SES schools or districts) were also collected.