580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can Item Response Tree models capture complex response processes beyond traditional IRT outcomes?
This research theme explores advances in Item Response Theory (IRT) that model the internal cognitive or psychological decision processes influencing item response selection. Beyond assessing terminal item responses, item response tree models characterize sequential, nested, and multidimensional decision-making pathways. This detailed modeling offers nuanced insights into psychological assessments, response omissions, and the structure of Likert-type scale responses, addressing limitations of classical IRT models that treat responses as flat outcome categories.
Key finding: Jeon et al. (2015) introduce a generalized item response tree (IRT) model that flexibly incorporates node-specific parametric forms, dimensionality, and covariates, allowing models to capture complex decision processes in... Read more
2. What advantages does Item Response Theory offer over Classical Test Theory in psychological test development and measurement precision?
This research area investigates the methodological and practical benefits of Item Response Theory (IRT) compared to Classical Test Theory (CTT) in psychological and educational assessments. It focuses on how IRT models provide invariant item and person parameters, permit precise measurement precision quantification at varying trait levels, and support refined test development practices. These advantages are crucial for improving test validity, reliability, and interpretability, especially in scales with graded responses.
Key finding: This study applies the graded response model (GRM) of IRT to positive and negative affect scales, demonstrating that IRT estimates person abilities invariantly across different test forms and reveals item discrimination... Read more
Key finding: This paper articulates fundamental measurement issues in psychology and marketing research that IRT can address more effectively than CTT, such as balancing reliability and construct validity and handling item wording... Read more
Key finding: This paper outlines the advantages of IRT in educational and psychological test development, emphasizing its probabilistic modeling of item responses as functions of latent traits and its ability to address inherent... Read more
3. How can item-fit and model-data fit be accurately assessed in IRT to identify aberrant items and improve measurement validity?
This theme focuses on the development and evaluation of statistical methods for assessing the fit of IRT models at the item level, crucial for ensuring accurate parameter estimation and valid test scores. It compares chi-square and entropy-based techniques, investigates challenges with traditional fit statistics due to model dependency and sample-specific grouping, and explores computational innovations to provide more precise diagnostics of item misfit, enabling enhanced item selection and test calibration.
Key finding: This study compares several item-fit statistics including EMRj, traditional chi-square (X2), likelihood ratio (G2), S-X2, and PV-Q1 through Monte Carlo simulations mimicking item-level misfit scenarios under a 2PL IRT model.... Read more
4. What computational methods and software can enhance IRT parameter estimation for complex models and simulation studies?
This area addresses the challenges of flexible IRT model estimation using Bayesian methods and computational resources. It covers implementation strategies using BUGS-language software for various common and extended IRT models, enabling customization for longitudinal or multi-level data structures. It also examines automation with R scripting to conduct large-scale simulation studies using stand-alone software packages, streamlining iterative model fitting and fit metric extraction critical for psychometric research.
Key finding: The paper presents detailed Bayesian modeling code in BUGS language for fitting common IRT models including the 2PL, 3PL, graded response, generalized partial credit, testlet, and generalized testlet models. It highlights the... Read more
Key finding: Lee demonstrates methods to automate complex and large-scale IRT simulation studies by leveraging R's scripting capabilities to generate datasets, prepare software inputs, invoke stand-alone IRT estimation tools (like... Read more
5. How can IRT be extended and applied to continuous and polytomous response data while accounting for measurement constraints?
This theme investigates the modeling of non-dichotomous responses in IRT, including continuous measurements such as response times and Likert-scale data. It explores latent trait models that incorporate distributional restrictions (e.g., response boundedness), extend traditional discrete IRT to continuous domains, and develop threshold models suited for polytomous items. Addressing response scale properties improves model appropriateness, measurement validity, and the treatment of response patterns in diverse assessment contexts.
Key finding: The article establishes a general IRT framework for continuous response variables that models responses as functions of latent traits while explicitly accommodating restrictions such as bounded or positive supports. It shows... Read more
Key finding: This study applies Nonparametric Item Response Theory (NIRT) methods, specifically the Mokken and Dominance Models, to examine the one-dimensionality and invariant item ordering assumptions of the TOEFL iBT listening test.... Read more