Abstract
Chetty, Friedman, and Rockoff (2014)[CFR] evaluate the degree of bias in teacher value-added (VA) estimates using a “teacher switching” research design, regressing changes in mean test scores across cohorts on changes in mean teacher VA. Recent studies (Kane, Staiger, and Bacher-Hicks 2014, Rothstein 2014) have found that regressing changes in mean scores in the prior grade on changes in mean VA also yields a positive coefficient, a result we confirm in our data. At first glance, this “placebo test” seems to suggest that CFR’s quasi-experimental teacher switching design is invalid: how can changes in current teacher quality affect students' past test scores? The problem with this placebo test is that teacher value-added is itself estimated from prior test scores. Hence, changes in current teacher VA will be mechanically correlated with changes in prior scores even when a research design is valid. We show that simple variants of the lagged score placebo test that remove this mechanical effect uncover no evidence of a relationship between changes in teacher VA and lagged test scores. We conclude that (1) the correlation between changes in teacher VA and prior test scores is not an informative placebo test and (2) the original CFR research design and results are valid.