Panel Paper: Methods for Accounting for Co-Teaching in Value-Added Models

Friday, November 9, 2012 : 8:40 AM
Hanover B (Radisson Plaza Lord Baltimore Hotel)

*Names in bold indicate Presenter

Heinrich Hock and Eric Isenberg, Mathematica Policy Research

As higher quality roster data becomes available, value added models of teacher effectiveness must be adapted to apportion responsibility for student achievement growth when a student is taught by multiple teachers. Such “co-teaching” can be fairly common and arises due to team teaching, re-grouping of students, student mobility, and/or teacher mobility. To estimate teacher effectiveness in value-added models with co-teaching, we consider three methods: the Partial Credit Method, Teacher Team Method, and Full Roster Method. The Partial Credit Method apportions responsibility between teachers according to the fraction of the year a student spent with each. In practice, however, when teachers have many students in common but teach few students individually, near-collinearities among the teacher measures may result in unstable value-added estimates. The alternative methods, which require fewer assumptions about the relative effectiveness of a teacher with solo-taught and shared students, do not suffer from near-collinearity, but also do not allow inference about the distinct contribution that individual teachers make to shared students. The Teacher Team Method uses a single record for each student and a set of variables for each teacher or group of teachers with shared students. The Full Roster Method contains a single variable for each teacher, but multiple records per student for shared students. We explore the theoretical properties of these two methods and then compare the estimates generated using student achievement and teacher roster data from a large urban school district. We find that both methods produce very similar point estimates of teacher value-added. However, the Full Roster Method better maintains the links between teachers and students and can generally be more robustly implemented in practice.