EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Robust-Mean–Geometric-Mean and Robust Haberman Linking with Invariant Item Discriminations Under Sparse Differential Item Functioning

Alexander Robitzsch

Mathematics · 02.05.2026 · 14 (9) · 1549

DOI: 10.3390/math14091549 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mathematics, online erschienen am 02.05.2026. Verfasst von Alexander Robitzsch. Die Arbeit nennt 66 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Comparison of two or multiple groups based on dichotomous items is a central task in item response theory (IRT) linking. This article considers the two-parameter logistic scaling model under sparse differential item functioning (DIF) in item intercepts and DIF-free item discriminations. Robust-mean-geometric-mean (RMGM) and robust Haberman (RHAB) linking are compared across several loss functions and under scaling models with noninvariant or invariant item discriminations. Two simulation studies show that invariant item discriminations improve the precision of estimated group means. In addition, the L0 loss function is generally preferable to the L1 and L0.5 loss functions when DIF proportions or sample sizes are large. Several empirical examples illustrate the proposed specifications.

Zitieren

APA

Robitzsch, A. (2026). Robust-Mean–Geometric-Mean and Robust Haberman Linking with Invariant Item Discriminations Under Sparse Differential Item Functioning. Mathematics, 14(9), 1549. https://doi.org/10.3390/math14091549

Vancouver

Robitzsch A. Robust-Mean–Geometric-Mean and Robust Haberman Linking with Invariant Item Discriminations Under Sparse Differential Item Functioning. Mathematics. 2026;14(9):1549. doi:10.3390/math14091549

BibTeX
@article{Robitzsch2026,
  author = {Robitzsch, Alexander},
  title = {Robust-Mean–Geometric-Mean and Robust Haberman Linking with Invariant Item Discriminations Under Sparse Differential Item Functioning},
  journal = {Mathematics},
  year = {2026},
  volume = {14},
  number = {9},
  pages = {1549},
  doi = {10.3390/math14091549}
}
Kostenlos starten →