[3M2-GS-10u] AI application

11:45 AM - 12:00 PM JST(2:45 AM - 3:00 AM UTC)

[3M2-GS-10u-04] An Equating Method Using a Biased Anchor Set in Item Response Theory

〇Yuta Kawasaki¹, Takeaki Sakabe¹, Keiichi Namikoshi¹, Yuko Sakurai¹, Satoshi Oyama^2,3 (1. Nagoya Institute of Technology, 2. Nagoya City University, 3. RIKEN AIP)

Keywords:

Item Response Theory,Equating,Bias Correction

This paper aims to enhance the stability and validity of equating in Item Response Theory (IRT). Equating links scales across different groups using common items (anchors). However, when the anchor set is selected from a specific region of the underlying distribution, equating accuracy may deteriorate. To address this issue, we propose a density ratio weighting method incorporating a reliability index to mitigate instability caused by small sample sizes. Experiments demonstrate that our method achieves higher accuracy than conventional approaches, with its effectiveness differing between pitcher and batter settings.