RMME/STAT Joint Colloquium

Assessment of Fit of Item Response Theory Models: Full-information and Limited-information Methods, Item and Person Fit Analysis, and Beyond

Dr. Sandip Sinharay

Educational Testing Service (ETS) Research Institute

Friday, October 11, at 11:15 AM ET

AUST 110

https://tinyurl.com/rmme-Sinharay

Item response theory (IRT) is one of the central methodological pillars supporting many large and high-profile assessment programs globally. IRT analysis is essentially a type of discrete multivariate analysis and is performed using IRT models that are latent variable models for discrete data. However, IRT models involve multiple assumptions like conditional independence, monotonicity etc. and the results obtained from IRT models may not be accurate if one or more of the assumptions are not met, that is, if there is IRT model misfit. This presentation will include a comprehensive review of the literature on the assessment of fit of IRT models. The presenter will discuss various approaches and concepts regarding IRT model fit including full-information and limited-information methods, residual analysis, item and person-fit analysis, Bayesian methods, analysis for differential item functioning, and assessment of practical significance of misfit. A real data example will be used to illustrate some of the approaches. One goal of the presentation is to stimulate discussions involving the audience members regarding IRT model-fit assessment.

 

---