An Investigation of the Sensitivity of Value-Added Estimates to Choice of Scale and Model

• In Event: 65.079 - Roundtable Session 23
  In Roundtable Session: 65.079-15 - Issues in Value-Added Modeling

Abstract

Many states employ value-added models (VAMs) in the evaluation of schools and educators. One persistent concern has been the sensitivity of the results to choice of model. In this study, utilizing data from the ECLS-K, two different VAMs were applied to two different representations of student progress, based on the same raw data. One is new to the literature and operates on Markov transition matrices; the other is a version of the Sanders model and employs IRT-scaled test data. Although the results were based on a single cohort, with small samples in each unit, there was reasonable consistency between methods both with regard to correlations and agreement in the tails of the distributions of the unit value-added scores.

Authors

• Henry I. Braun, Boston College

• Catherine S. Trapani, WestEd

• Yanxuan Qu, ETS