Effects of Unmodeled Latent Classes in Value-Added Modeling

• In Event: Roundtable Session 18
  In Roundtable Session: Assessing Student Growth: Methods, Issues, and Applications

Abstract

Omitting student growth patterns over time in value-added modeling (VAM) eliminated findings that positive level 2 effects led to positive bias (better teachers look even better) while negative level 2 effects led to negative bias (worse teachers look even worse). Effective VAM requires multilevel growth mixture model rather than multilevel model.

Authors

• Futoshi Yumoto, IMPAQ International

• Rochelle Tractenberge, Georgetown University

• Gregory R. Hancock, University of Maryland

• Robert J. Mislevy, Educational Testing Service