Value-added models of teacher effectiveness yield consistent estimates of teacher
quality under the assumption that students are randomly assigned to classrooms
conditional on ability. Rothstein (2010) tested and rejected this underlying sorting
assumption, casting doubt on the usefulness of the value-added framework.
In this paper, I illustrate that the falsification tests employed by Rothstein perform
quite poorly in small samples and I propose an alternative testing strategy that
is less sensitive to sample size. I also show that none of the proposed falsification
tests works well when the achievement production function is misspecified.
Finally, I return to the same North Carolina sample employed by Rothstein and
retest the assumption of conditional random assignment. Once I account for the
“smallness” of the data and allowteacher inputs to persist at reasonable rates, I fail
to reject conditional random assignment.
Keywords. Teacher value-added, model testing.
JEL classification. I20, C10.