I’m a very strong proponent for making educated decisions based on statistics. I recently read an article from the NEA magazine on “Value-Added” measures for teachers. In short what I know about this system, assigns a value to a teacher based on specific factors that the district finds influential. I’m sure the school district uses factors that are signficant in their models like socio-economic status of students, race, content area, etc. The NEA magazine of course dismisses any merit in this research and this is confirmed by a paper published from Harvard University by Rothstein. He argues that prediction of student performance from early grades can be predicted by later grades. The argument takes realistic implications from administration assigning more difficult students to specific teachers and other similar scenarios. I believe the system could be much more improved upon to actually make much better prediction measures for teacher performance. It’s actually very hard to find how this “Value Added” system is calculated, the NEA magazine says that it’s “secret”. I suppose this secrecy is to protect teachers and administration of exploiting the system. I think the whole problem of this system lies in the fact that when we perform predictive trends, surfaces, etc. we are trying to minimize our error in prediction. A teacher this year who scores above average and deserves a pay raise should next year have the exact same probability of having a pay deduction. The errors from the prediction should be uncorrelated and hopefully random. To find a teacher who actually performs above the standards would produce an incorrect model and prediction. The whole statistic is to make teachers fail or not receive merit based pay. There would be a random process in assigning teachers payment which is not merit based at all. I believe a much better approach to solving the statistics in a problem like this is to expect autocorrelation between students from year to year. This is to basically say that a teacher in grade sevens performance should only be created from the student’s prior observations.
Lastly, let me make it clear that I think incentive based pay is a great idea from people who have never worked in public education. There are multiple factors that play a part in many students and those same factors may not play a part in others. To try to generalize a population to a specific model is ridiculous would always fail. We could minimize these failures, but we would in essence “Leave Children Behind”.