Generalizability In Item Response Modeling

Collaborating Institutions: 
School of Education, University of Colorado, Boulder

School of Education, University of Colorado, Boulder Participants: 
Derek C. Briggs (Co-PI)

BEAR Center Participants: 
Jinnie Choi, Mark Wilson (Co-PI)


The Generalizabilty in Item Response Modeling (GIRM) project aims to find a way to integrate two psychometric approaches: Generalizability Theory (GT) and Item Response Theory (IRT). The GIRM approach incorporates the sampling model of GT into the scaling model of IRT by making distributional assumptions about the relevant measurement facets. In Briggs & Wilson (2006), it is shown that it is possible to estimate GT variance components concurrently with traditional IRT parameters, by specifying a random effects measurement model and using Markov Chain Monte Carlo (MCMC) estimation methods. The GIRM project focuses on the extension of the GIRM approach to a wide range of other potential measurement conditions.


Currently, the focus is to expand the GIRM approach to more complex designs, which include (1) between-item multidimensionality, (2) within-item multidimensionality, and (3) polytomous item responses. Major tasks include building comparable theoretical framework for both GT and IRT for the particular design, specifying appropriate explanatory item response models (EIRMs; De Boeck & Wilson, 2004) along with possible combinations of distributional assumptions, programming with WinBUGS and R software, conducting simulation studies to explore the properties of the approach, and conducting empirical studies to explore possible applications. The outcomes are expected to contribute to the methodological advance of the field of educational measurement.