Comparing Joint and Chain IRT Linking Under Random Item Parameter Drift Across Many Time Points

Read full text Summarise with VIRAT
Authors: Oskar Engels, Oliver Lüdtke, Alexander Robitzsch
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This study compares four IRT linking methods across seven time points under random uniform item parameter drift (IPD): concurrent calibration, Haberman, Haebara, and Stocking-Lord linking. Each method is implemented under three linking approaches: joint (all time points simultaneously), chain (consecutive time points sequentially), and restricted joint (only time points 1 through t when estimating at t). A simulation study evaluates bias and RMSE of the distribution parameter estimates. Without IPD, most methods produce satisfactory bias, with concurrent calibration and Stocking-Lord joint linking as the most efficient. Under IPD, only Haberman linking yields unbiased estimates across all conditions and linking approaches. The other methods show increasing bias as IPD magnitude and the number of time points grow. Under large IPD, joint linking reduces bias for concurrent calibration, whereas chain linking reduces bias for Haebara and Stocking-Lord; under moderate IPD, chain linking also reduces bias for concurrent calibration. Restricted joint linking produces intermediate estimates for concurrent calibration and, to a lesser extent, for Stocking-Lord. For Haebara, variant choice substantially affects bias under IPD; for Stocking-Lord, it has little impact. An empirical application to longitudinal mathematics achievement data illustrates these findings.

Fields of Study & Subjects

Psychometric Methodologies and Testing Educational Assessment and Pedagogy Statistical Methods and Bayesian Inference

Publication details

Journal / source
PsyArXiv (OSF Preprints)
Publication date
28 March 2026
Language
EN
Document type
Preprint