出版社：The Japanese Association of Educational Psychology
摘要：本稿では，共通受検者計画を用いて連鎖的等化を実施する場合の等化誤差とテスト結果の実用性について，モデル不適合の観点から検証した。モデル不適合の要因として,共通受検者集団の能力分布の非正規性と特異項目機能を取り上げた。推定値のバイアス，標準誤差，受検者の段階評価への分類精度を評価指標，共通受検者集団の能力分布の変化（平均変化条件，歪度変化条件，固定条件），等化の連鎖回数（最大連鎖数＝5回），特異項目機能の種類と程度（均一DIFと不均一DIF/効果量の大小）の3要因を独立変数とした上で，各要因の影響を検証するためにシミュレーション実験を実施した。実験の結果，（a）尺度得点の推定値の分散の約99％は偶然誤差によって説明される，（b）DTM （Difference That Matters） 基準の下で，3要因は実用上，問題となるバイアスを生じさせない，（c）連鎖数は標準誤差に対して最大の影響力を持ち，連鎖数が4以上となる場合，DTM基準を超える標準誤差が生じる可能性が高くなる，（d）受検者の段階評価への分類精度に対する3要因の影響は約0.1％程度に限定される，といった示唆が得られた。.
其他摘要：In the present study, a common examinee design was used to verify the equating error and practical consequences of the chained equating method, from the viewpoint of model misfit. The factors of model misfit that were selected were (a) the non-normality of the ability distribution of common examinees, and (b) differential item functioning (DIF). The independent variables were the change in the ability distribution of the common examinees, the number of equating chains, and the type and degree of differential item functioning. The evaluation criteria were the bias and standard error of the estimated scaled scores and the accuracy of classifying the examinees into 4 categories. A simulation study examined influences of the 3 independent variables. The results indicated the following: (a) 99% of the variance of the estimated scaled scores was random error, (b) using a DTM (difference that matters) criterion, the 3 factors did not cause a bias that was problematic in practice, (c) the number of chains had the greatest influence on the standard error, so that if the number of chains was 4 or more, a standard error exceeding the DTM criterion was likely to occur, and (d) the accuracy of classifying the examinees into the 4 categories was hardly affected by the 3 factors.
