Education types, at least those who live and work inside the beltway, have been buzzing about growth models for quite some time, and one type of growth model, value-added, has been especially intriguing to policymakers.

As a 2004 Rand report put it:

Policymakers see VAM as a possible component of education reform through improved teacher evaluations or as part of test-based accountability. They are particularly intrigued by VAM because of the view that its complex statistical techniques can provide estimates of the effects of teachers and schools that are not distorted by the powerful effects of such noneducational factors as family background. 

The RAND report went on to caution that VAMs were a little rough around the edges, and economist Dale Ballou noted VAMs' limitations and problems way back in 2002.  

Recent work by Princeton researcher Jesse Rothstein casts even more doubt, adding an element of the absurd.  Using an ingenious approach, Rothstein argues that supposedly reliable value-added models can result in impossible, time-warping correlations. (To my mind, this is akin to the indirect proofs I learned --and later taught -- in geometry class. Start with an assumption; take it where it leads you; if it leads to a contradiction, the assumption is false.)

Rothstein writes: 

My tests are based on a simple falsification exercise: Future treatments cannot have causal effects on current outcomes, and models that indicate such effects must be misspecified.  I demonstrate that a simple VAM of the form typically used in the literature indicates large effects of current teachers on past achievement:  A student's 5th grade teacher has nearly as large an effect on her 4th grade achievement growth as on her learning during 5th grade.  This is direct evidence of non-random assignment.

If 5th grade teachers have an affect on 4th-graders' performance, then either time travels backward or there's something wrong with the model.

Or maybe there's something wrong with Rothstein's work.  If so, he'll hear about it.  He's inviting comments on his findings, currently in the form of a working paper.

Via Eduwonkette. 

*The Center for Public Education has a handy "At a Glance" piece on growth models here.