Robin Hanson sends along some commentary that simply can’t be ignored.

Merit pay for teachers is an idea that is almost 100 years old and has been subject to much research.  In one study conducted in 1918, “48 percent of U.S. school districts sampled used compensation systems that they called merit pay.” … The evidence shows that merit-pay plans seldom last longer than five years and that merit pay consistently fails to improve student performance.  … [Researchers] also showed that cheating [by teachers] was quite sensitive to the size of the incentives provided for enhancing student scores.  … The same problems emerged when merit-pay systems were implemented in the 1980s. … “It is like policy makers suffer from amnesia.” (pp.22-24) …

The evidence strongly suggests that students learn better when they are not graded and certainly not when they are graded on a curve.  … When drill instructors were tricked into believing that certain randomly selected soldiers would achieve superior performance, those soldiers subsequently performed far better on tasks like firing weapons and reading maps.  (p.38)

Ending social promotion harms students and schools, and the strongest negative effects are found in the best, most rigorous studies.  At least 55 studies show that when flunked students are compared to socially promoted students, flunked students perform worse and drop out of school at higher rates.  One of the most careful studies found that, after controlling for numerous alternative explanations including race, gender, family income, and school characteristics, students held back one grade were 70 percent more likely to drop out of high school.  (p.51)

I don’t know the authors and I don’t know the studies but you this has to peak interest on this issue.

Especially interesting that students reportedly perform better when they are not graded. Its hard not to think that selection effects are not at work here. I just can’t picture the student model which produces this, even given a really large “nervousness” effect that degrades performance.