This article caught my mind while looking for something else. Houston has decided to reward merit by paying teachers more if their students do well in tests:
The pay incentives are to be based on three components, or "strands."
One will reward teachers based on how much their school's test scores have improved compared with the scores of 40 other schools with similar demographics around the state. Another will compare student progress on the Stanford 10 Achievement test and its Spanish-language equivalent to that of students in similar classrooms in the Houston district. The third measure will be student progress on the statewide Texas Assessment of Knowledge and Skills test, as compared with that in similar Houston classrooms.
About half the district's teachers will be eligible for stipends in all three categories, for a total of $3,000. The system's 305 principals with the best-achieving teachers could earn as much as $6,000 in merit pay, and the 19 executive principals and five regional superintendents will be eligible for up to $25,000.
The linked article discusses many negatives with this plan: it encourages teaching the test only, it doesn't reward Special Education teachers or art teachers. But the article doesn't really describe the worst aspect of the scheme: Just imagine if we decided to do the same with physicians' pay. Pay physicians more if their patient's vital readings improve.
It's easier to see the problems in tying the reward to the "results" in that case, and the two cases have the same problems. Both patients and students enter the system with varying abilities to improve, and the outcomes for both patients and students depend not only on the doctor or the teacher but also on themselves and their families.
What this system will reward is teaching the cream-of-the-crop students, and it will give teachers an incentive to try to get rid of poor students if at all possible. It will also reward those that learn to game the system, by cheating, if necessary.