Two-Step Residual-Averaging Value-Added Estimation
RCTClinical Trial
Directly estimating teacher effectiveness across large multi-district datasets with diverse student populations requires a computationally tractable method to strip away student baseline traits before aggregating performance to the instructor level.
Picture this
Evaluating employee performance by first calculating how much each employee outperformed their personal benchmark given their specific job assignment, and then averaging those extra performance points across all projects completed by that employee over the year.
What the evidence says
Averaging student residuals provided an unbiased baseline-adjusted measure of teacher performance that, when combined with shrinkage adjustments, predicted post-randomization achievement gains with an IV slope of 0.955 (SE = 0.123).
- Who was studied
- N = 31,439 randomized students and 1,181 randomized teachers across 6 urban public school districts.
- How
- Two-step regression modeling: Step 1 regresses standardized end-of-year test scores on student baseline test scores, demographics, and classroom mean covariates; Step 2 averages individual student residuals $(\epsilon_{it})$ by teacher and subject to form $\hat{\tau}_{jt}^{S}$.
What to do
Calculate student-level achievement residuals from a first-stage covariate regression, then average the residuals by instructor to construct raw value-added estimates.
From the source
"The residuals from the model above represent the degree to which each student outperformed or underperformed similarly situated students. By averaging the residuals across a teacher's students, we used these to generate teacher-level value-added estimates ($\hat{\tau}_{jt}^{S}$) for a teacher in each subject."
Have We Identified Effective Teachers? Validating Measures of Effective Teaching Using Random Assignment