This week’s @nberpubs includes my working paper with @mpbitler @ThadDomina and @EmilyKPenner. We apply value-added models commonly used in practice to estimate teacher “effects” on student height. We find these effects are about as large as those on achievement (1/n)

The good news: these “effects” don’t seem to reflect sorting or selection bias. That we find teacher effects on height does *not* invalidate a large body of research that finds systematic and important effects of teachers on students. https://www.nber.org/papers/w26480  (2/n)

Those studies have gone to great lengths to validate teacher VAMs in other ways, including documenting correlations with other measures of effectiveness (a la the Met project), random assignment, and quasi-experiemental tests using teachers who change schools. (3/n)

In other words, it is possible *both* for us to find nonsensical teacher effects on height, and for there to be real and large effects on achievement. (4/n)

The bad news: thanks to noise (small samples), random and fixed effects models will find “effects” where there clearly are none. We think this was a clever way to illustrate this point, and suspect there are other literatures where this issue comes up. (5/n)

There are methods for “shrinking” VAMs to separate “signal” from “noise”, but (as we discuss in the paper) it’s not clear these methods are that helpful in educational practice where states and districts use VAMs to evaluate teachers in real time. https://www.nber.org/papers/w26480  (n/n)