We compare two alternative methods to account for the sorting of students into academic tracks. Using data from an urban school district, we investigate whether including track indicators or accounting for classroom characteristics in the value-added model is sufficient to eliminate potential bias resulting from the sorting of students into academic tracks.
We find that accounting for two classroom characteristics -- mean classroom achievement and the standard deviation of classroom achievement -- may reduce bias for middle school math teachers, whereas track indicators help for high school reading teachers. However, including both of these measures simultaneously reduces the precision of the value-added estimates in our context. In addition, we find that while these different specifications produce substantially different value-added estimates, they produce small changes in the tails of value-added distribution.