Using Multiple Dimensions of Teacher Value-Added to Improve Student-Teacher Assignments

• In Event: Unpacking Value-Added Measures of Teacher Performance: Implications for Disadvantaged Students and the Parents and Teachers Who Support Them

Abstract

Value-added models that use student test performance to measure teacher effectiveness are becoming increasingly prevalent in the evaluation of teachers, policies, and programs. However, current uses of value-added modeling largely ignore or assume away the potential for teachers to be more effective with one type of student than another. Similarly, it is not hard to imagine that elementary school teachers, who typically teach all subjects, may be more effective in one subject than another. The consistency of value-added measures is important given their increased use in evaluation, but furthermore, if teacher quality does indeed vary by student type or subject, there is potential to raise student achievement through optimally matching students to teachers best suited to teach them.

Using eight years of rich, administrative data from a large, urban district, I estimate value-added models that allow teacher effectiveness to vary by student type or subject. I then simulate different sorting procedures, using the estimated teacher value-added to quantify the achievement gains that accrue from such sorting. In doing so, this paper examines four research questions:

1. Does teacher effectiveness, as measured by value-added, vary within a teacher by whether students are low or high ability, as measured by prior test scores? Does it vary by student gender?
2. To what extent would re-sorting students by optimally matching them to teachers best suited to teach them result in increased achievement?
3. Does teacher effectiveness vary within a teacher by subject, specifically math and reading?
4. To what extent would assigning teachers to the subject in which they have a comparative advantage in teaching result in increased student achievement?

The findings indicate a high level of stability of teacher effectiveness across gender and ability, with all value-added correlations greater than .9. The correlation between teacher effectiveness in math and reading is slightly lower at .72. The lower the correlation, the more leverage there is for producing gains through the re-assignment process I carry out. Average overall gains range from .002 to .017 for the student subgroups across math and reading. The simulation where teachers are assigned to classrooms based on their respective comparative advantages in subject taught reveal overall gains of .029 using unstandardized value-added scores and .016 using standardized scores.

My findings indicate that a good teacher is a good teacher in most contexts, though there are some differences, especially by subject taught. Though the simulations in this paper are unlikely to yield the true gains that could be accrued from the re-sorting procedures, they utilize the current labor force and do not rely on assumptions about longer term labor market shifts to improve student achievement. Furthermore, all of the data used in this paper comes directly from the district, so schools wanting to increase student achievement by more optimally matching students and teachers could consider the strategy employed here. As such, the paper provides policy-relevant evidence about the potential benefits of using multiple dimensions of teacher value-added measures to more strategically create student-teacher assignments.

Author

• Lindsay A. Fox, Stanford University