Email Bookmark and Share

Invited Symposium

Modern Psychological Measurement: New Developments in Item Response Theory

Sunday, May 24, 2009, 10:30 AM - 11:50 AM
Yerba Buena 3 - 4

Li Cai Chair: Li Cai
University of California, Los Angeles

Item Response Theory (IRT), already widely used in large-scale educational assessment, is now increasingly applied to measure psychological constructs in research settings. A notable example is the recent surge of interest among mental health researchers in the application of IRT and its close relative – Computerized Adaptive Testing – to develop standardized measurement instruments for patient reported outcomes. This change of focus, from educational assessment to psychological measurement, has challenged traditional IRT methods and provided opportunities for new developments. As a result, the past few years have seen exciting research on 1) new IRT models, 2) new methods for estimation, and 3) improved software implementation. This symposium draws on the expertise of speakers who are at the forefront of each of these three areas, and provides a forum for cutting-edge research in quantitative psychology.

Michael C. Edwards

Potential Applications for Multidimensional Item Response Models
Michael C. Edwards
Ohio State University
The past decade has seen a tremendous amount of progress in the estimation of multidimensional item response theory (MIRT) models. Although these models have existed conceptually for quite some time, estimating parameters for these models was not an easy task. Solutions existed, but they were generally hard to find, highly constrained, difficult to use, or all of the above. Despite these limitations, there never seemed to be a great outcry from the research community for more flexible MIRT models. Computer-intensive estimation procedures have overcome the computational barriers and made MIRT models feasible (and flexible). Now that we have access to this class of models we return to the basic question: What do we do with them? In this talk I begin by providing some basic categories into which MIRT models fall. I will also review and introduce new uses for various sorts of MIRT models. It is my hope that, by discussing some beneficial uses of MIRT models, researchers will begin to see how MIRT models may be applicable in their own research.

Carol M. Woods

IRT-based Differential Item Functioning Testing with a Non-Normal Latent Variable
Carol M. Woods
Washington University in St. Louis
This research describes an implementation of item response theory based likelihood ratio tests of differential item functioning (IRT-LR-DIF) wherein the latent density is presumed standard normal for the reference group but estimated from the data for the focal group. The new procedure, RC-DIF, is designed for types of variables that are approximately normal for the majority or mainstream group with which the test was constructed, but possibly not for some focal groups. When the focal group density is non-normal, RC-DIF provides more accurate results than standard IRT-LR-DIF (for which both densities are presumed normal).

Li Cai

Flexible Multidimensional, Multiple Categorical IRT Modeling
Li Cai
University of California, Los Angeles
I will introduce a multiple-group, confirmatory, multidimensional, multiple categorical IRT modeling framework. The present framework allows arbitrary user-defined restrictions on the item parameters so that competing measurement models about the underlying psychological process can be tested. Building on existing frameworks for mixing unidimensional multiple categorical IRT models offered in software such as MULTILOG, the present framework can handle, in a single analysis, any arbitrary combination of multidimensional dichotomous, graded, or nominal items, in one or more groups. The flexibility and efficiency afforded by the new modeling framework will be demonstrated by empirical examples.

David M. Thissen

David M. Thissen (Discussant)
University of North Carolina at Chapel Hill

Go back

Intelligence and How To Get It: Why Schools and Cultures Count - Richard E. Nisbett