Elementary School Data Issues for Value-Added Models: Implications for Research

Abstract

Researchers often presume that it is better to use administrative data from grades 4 and 5 than data from grades 6 through 8 for conducting research on teacher effectiveness that uses value-added models because (1) elementary school teachers teach all subjects to their students in self-contained classrooms and (2) classrooms are more homogenous at the elementary school level. We examined the first issue by using data on teacher–student links in which teachers of mathematics and/or English/language arts had verified the subjects and students they taught. We compared these data to teacher–student links from the original administrative data. Results show that instruction is often departmentalized in these grades. About one in six elementary school teachers in the original data was linked to a subject that he or she did not teach. To examine the second issue, we computed the variation in baseline student achievement within classes, between classes at the same school, and between schools. We found more within-school variation in pretest scores in middle school grades but an offsetting amount of between-school variation in upper elementary grades.

Keywords:

• Value added
• roster confirmation
• teacher

Notes

This paragraph describes the roster confirmation process in the 2010–2011 and 2011–2012 school years, the years of the data used in Sections III and IV.

A small number of differences may be explained by teacher mobility, since the unconfirmed rosters were compiled in the fall and the confirmed rosters in the spring. Some teachers may have moved in and out of DCPS during the school year. A few mismatches may also have occurred because the unconfirmed rosters included only teacher names (but not teacher IDs), so we may have inadvertently failed to match a few teachers who were present in both rosters, such as married women who changed their last names.

We excluded grades in schools with a single classroom because it was not possible for a school to create separate tracks in this case. Instead, because there can be no variation between classrooms in these schools, we included them in the results by assigning a value of 0 for between teacher/class variation and weighting the HLM estimates by one minus the share of students who were excluded. This was a substantial share of schools in DCPS: 19% of school–grade combinations had one classroom in mathematics and 20% had one classroom in reading. If we weight each school–grade combination by the number of students, 9% of school–grade combinations had one classroom.