American University
Browse
auislandora_72353_OBJ.pdf (331 kB)

Should value-added models control for student absences?

Download (331 kB)
online resource
posted on 2023-08-04, 06:20 authored by Seth GershensonSeth Gershenson

Whether or not value-added models should control for contemporaneous student absences is theoretically ambiguous, as such absences are only partly outside of teachers’ control. Teachers often feel strongly that value-added models should account for student attendance, and many districts’ value-added models condition on lagged student absences as a result. Using matched teacher-student administrative data from a state-wide longitudinal data system, this note investigates the practical importance of this modeling decision for value-added measures of teacher effectiveness (VAMs). This is done by comparing VAM-based rankings of teacher effectiveness generated by value-added models that either control for current absences, control for lagged absences, or exclude student absences altogether. Regardless of how between-school differences are accounted for, VAM-based rankings of teacher effectiveness are insensitive to how, and whether, student absences enter the value-added model’s conditioning set. Spearman Rank Correlations are always larger than 0.99 for both math and reading VAMs, suggesting that whether or not value-added models control for annual student absences is a relatively unimportant modeling decision. These results are consistent with recent research suggesting that simply conditioning on lagged achievement yields approximately unbiased VAMs. Moreover, these findings suggest that controlling for student absences in teacher evaluation systems’ value-added models is a relatively inexpensive way to increase teacher buy-in.

History

Publisher

American University

Handle

http://hdl.handle.net/1961/auislandora:72353

Usage metrics

    Public Administration & Policy

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC