Working Paper: NBER ID: w19423
Authors: Raj Chetty; John N. Friedman; Jonah E. Rockoff
Abstract: Are teachersʼ impacts on studentsʼ test scores ("value-added") a good measure of their quality? One reason this question has sparked debate is disagreement about whether value-added (VA) measures provide unbiased estimates of teachersʼ causal impacts on student achievement. We test for bias in VA using previously unobserved parent characteristics and a quasi-experimental design based on changes in teaching staff. Using school district and tax records for more than one million children, we find that VA models which control for a studentʼs prior test scores exhibit little bias in forecasting teachersʼ impacts on student achievement.
Keywords: teacher quality; value-added; student achievement
JEL Codes: H0; H52; H75
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Controlling for prior test scores (C29) | VA models exhibit little bias (C52) |
| Omitting parent characteristics (D15) | forecast bias (C53) |
| Omitting twice-lagged scores (C22) | forecast bias (C53) |
| 1 standard deviation improvement in teacher VA (I24) | normalized test scores increase (C52) |
| Controlling for prior scores (C29) | correlation between VA estimates and parent characteristics disappears (C29) |