Measuring teacher performance

I have been thinking about the problem of how one might use student test results to help measure the performance of teachers. I just want to focus on one small element of the problem. I realize that it is very, very difficult to construct tests that measure the teacher's contribution to the students' ability to think analytically and creatively, or the teacher's motivation of students to embark on a vigorous course of lifelong learning. Indeed, I recognize that providing incentives to teachers who get their students to do well on tests that measure (or mis-measure) only part of the benefits of their courses may well bias their work and diminish the overall performance of the teachers (and students). I even realize how difficult it is to construct tests that effectively measure facts learned and techniques mastered by the students. But I just want to post on the quantitative approach to measuring teacher impact on student achievement.

If you assume that a teacher teachers something between 100 and 150 students in the course of a semester, then test results should provide some information about the teacher's performance. Of course, one should not assume that when students perform well on tests it is because their teachers did their job well that semester. Indeed, very good students from families that value education, in classes filled with equally good peers, who were taught well in previous years by other teachers may do quite well on a test after a semester taught by a very poor teacher.

How about an approach in which each student's test scores are predicted based on that student's previous test results and other variables independent of the current teacher, such as the scores of classmates on previous tests, the socio economic status of parents, and an indicator of the school facility quality. Then the teacher impact might be identified with the distribution of student test scores around their predicted values. Statistical tests could be applied to discover if the teachers students did better or worse on average than expected. Indeed, the distribution could be useful in discovering whether the teacher was favoring good students or students with weak performance, or indeed teaching to the average student and neglecting the needs of those on either end of the continuum.

My thinking was occasioned by a local story about a school system that is planning to provide bonuses to teachers it judges to have performed well based in part on student test results. In such a scheme, it would seem more fair to reward teachers whose students's results were (statistically) significantly better than expected.

There remain a bunch of issues. What happens if a teachers' students do well in one subject and poorly in another? Should success in bringing weak students up to average performance or success in bringing strong students to even greater success be more rewarded? Is success in mathematics and science worth more than success in civics and history? Indeed, the analysis of these issues might be more important in informing teachers and students about what the system feels to be important than the test based reward system itself.

There is also a psychological effect that comes from teachers knowing that their performance is being measured and that an attempt will be made to reward the best teachers, That effect may be beneficial even if the process of making those awards is faulty. Indeed, if one believes in the use of quantitative techniques to measure student and teacher performance, it makes sense also to use quantitative techniques to measure the value of the system used to reward high-performance teachers. One would hope that the average test results would go up with the system, as compared with previous test results and the results in control schools or national averages.