Teacher Value-Added in the Absence of Annual Test Scores: Utilizing Teacher Networks
Abstract: This paper proposes a novel methodology for estimating teacher value-added in the presence of non-random teacher-student sorting and the absence of annual standardized student test scores. Rather than relying on lagged student test scores to control for nonrandom teacher-student sorting on student ability, I exploit within-student across-subject variation in test scores and teacher "networks" - teachers in the same subject who teach groups of students observed with a unique teacher in another subject. The resulting estimates closely recover the true parameters of teacher value-added in Monte Carlo simulations and align well with estimates from standard methodologies in New York City data where lagged test scores are available. The methodology substantially expands the research on teacher value-added, as the majority of educational settings do not rely on standardized testing in consecutive grade levels. I apply the method to French middle school teachers and find that a 1 SD increase in teacher value-added within school improves student scores by 0.10 SD in Math and 0.07 SD in French. I show that using a "hybrid" methodology - which augments the network estimator to control for lagged scores - in settings where lagged scores are available can outperform standard methods under specific sorting patterns by accounting for sorting on time-varying student unobservables.
Presentations: Columbia University (2024), IZA/ECONtribute Workshop on the Economics of Education (2023), CESifo / ifo Junior Workshop on the Economics of Education (2023), 38th meeting of the European Economic Association (2023), SSE Quality in Education Conference (2023), 18th Doctorissimes Conference (2023), European Association of Labour Economists (EALE) Conference (2023), PSE Applied Economics Seminar (2023), ENS Workshop in Economics of Education (2022), PSE Labour and Public Economics Seminar (2022).