Examining the effects of automatic generated anchor items on equated parameters in NEAT equating

Abstract

This study explores the use of Automatic Item Generation (AIG) in equating, specifically focusing on its impact on equating outcomes when AIG items are used in the anchor set. The simulation study addresses how within-family variance of item siblings, group equivalence, and equating methods affect the equated item parameters, person parameters, and linking coefficients. Findings reveal that increasing within-family variance leads to greater error in equated parameters and linking coefficients. Additionally, differences in examinee group equivalence also increase error, particularly in the equated person parameters. The choice of equating method further influences error, with concurrent calibration and fixed item parameter estimation (FIPE) showing lower error compared to other methods. The study concludes that while using AIG item siblings in the anchor set is feasible, maintaining similarity in psychometric properties is crucial to minimize error. Concurrent calibration and FIPE are recommended for mitigating errors when within-family variation is present in a 2PL model.