Model Selection, Fit, and the Invariance Property in Item Response Theory

• In Event: Roundtable Session 14
  In Roundtable Session: Evaluating Measurement Invariance

Abstract

Many of IRT’s applications rely on the invariance property of the models. In order for the invariance property to hold, model assumptions, such as model fit, must be satisfied. An important, but often overlooked aspect of applying IRT is selecting an appropriate model to represent the probability of a correct response. Since many statewide assessments and certification programs use the 1PLM/Rasch model, it is important to examine whether applying the 1PLM provides acceptable results in terms of holding the invariance property when model data misfit exists. This study investigates the effects of applying the 1PLM to items representing responses to multiple-choice items by using empirical IRFs to simulate model misfit.

Authors

• MinJeong Shin, American Institutes for Research

• Craig S. Wells, University of Massachusetts - Amherst

• Jennifer Randall, University of Massachusetts