Using a multilevel random item Rasch model to examine item difficulty variance between random groups.

Autor/inn/en	Hartig, Johannes; Köhler, Carmen; Naumann, Alexander
Titel	Using a multilevel random item Rasch model to examine item difficulty variance between random groups.
Quelle	In: Psychological test and assessment modeling, 62 (2020) 1, S. 11-27 PDF als Volltext kostenfreie Datei Link als defekt meldenVerfügbarkeit
Sprache	englisch
Dokumenttyp	online; Zeitschriftenaufsatz
ISSN	2190-0507
Schlagwörter	Rasch-Modell; Mehrebenenanalyse; Methode; Vergleichsuntersuchung; Leistungsfähigkeit; Simulation
Abstract	In educational assessments, item difficulties are typically assumed to be invariant across groups (e.g., schools or countries). We refer to variances of item difficulties on the group level violating this assumption as random group differential item functioning (RG-DIF). We examine the performance of three methods to estimate RG-DIF: (1) three-level Generalized Linear Mixed Models (GLMMs), (2) three-level GLMMs with anchor items, and (3) item-wise multilevel logistic regression (ML-LR) controlling for the estimated trait score. In a simulation study, the magnitude of RG-DIF and the covariance of the item difficulties on the group level were varied. When group level effects were independent, all three methods performed well. With correlated DIF, estimated variances on the group level were biased with the full three-level GLMM and ML-LR. This bias was more pronounced for ML-LR than for the full three-level GLMM. Using a three-level GLMM with anchor items allowed unbiased estimation of RG-DIF.
Erfasst von	DIPF \| Leibniz-Institut für Bildungsforschung und Bildungsinformation, Frankfurt am Main
Update	2021/3