Working Paper: NBER ID: w19803
Authors: David J. Deming
Abstract: Value-added models (VAMs) are increasingly used to measure school effectiveness. Yet random variation in school attendance is necessary to test the validity of VAMs, and to guide the selection of models for measuring causal effects of schools. In this paper, I use random assignment from a public school choice lottery to test the predictive power of VAM specifications. In VAMs with minimal controls and two or more years of prior data, I fail to reject the hypothesis that school effects are unbiased. Overall, many commonly used VAMs are accurate predictors of student achievement gains.
Keywords: No keywords provided
JEL Codes: I21; I24; J24
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| VAMs with minimal controls (C36) | Student achievement gains (I24) |
| Adding more covariates (C39) | Predictive accuracy of VAMs (C52) |
| VAM estimates (C36) | Higher test scores for lottery winners (H27) |
| Prior test scores (C12) | Sorting across teachers (A14) |
| Lagged test scores (I24) | Value-added interpretation of school effects (I21) |
| Winning the lottery (H27) | Student achievement (I24) |