Entropy-Based Measures for Global and Local Model-Data Fit in Item Response Theory

• In Event: Modern Applications in Item Response Theory

Abstract

In this study we, (1) provide a generalized conceptual overview of data-model fit for model-, item-, and person-fit indices, (2) adapt entropy, entropy misfit, and the entropy misfit ratio (Ei, EMi, and EMRi, respectively) presented previously for person-fit in Item Response Theory (IRT) for person-, item- and model-fit in dichotomous IRT models, (3) present preliminary results from a simulation study to provide empirical evidence of how the fit indices function relative to other measures, and 4) provide confidence intervals (CI) derived from an empirical distribution to determine if entropy measures fit a theoretical distribution. We present initial findings of simulation work to fit model level data to theoretical distributions and continue work that investigates type I error and power.

Authors

• William R. Dardick, The George Washington University

• Brandi A. Weiss, The George Washington University