IRT真實分數等化於群體不變性假設之探究－以TASA 2007英語文題組測驗為例

Abstract

群體不變性是指不同次群體的測驗分數，經過相同的等化程序後，其轉換後之量尺分數應該相同。因此，為了要滿足測驗公平性的原則，等化在進行時必須符合群體不變性之需求（Kolen & Brennan, 2004）。近年來，許多研究皆在檢驗大型測驗進行量尺化程序後是否仍符合群體不變性的性質，例如：Yang與Gao（2008）透過IRT真實分數等化方法檢驗CLEP題組測驗的量尺分數是否符合群體不變性。此外，由於Yang與Gao使用Rasch測量模式估計題組測驗，而非使用題組反應理論模式。因此，本研究將以2007年臺灣學生學習成就評量資料庫英語文題組測驗資料為例，檢驗測驗資料是否滿足等化群體不變性的性質，並探究使用不同測量模式的等化效果。

In this study, population invariance refers to treating the test scores of different subgroups. Following identical equalization processes, the scaled scores after transformation should be the same. Thus, to ensure the fairness of tests, equalization must adhere to the demands of population invariance (Kolen & Brennan, 2004). In recent years, numerous studies have examined whether tests meet population invariance demands after scaling is performed. For example, Yang and Gao (2008) examined whether scaled CLEP scores using item response theory true score equating method conformed to the principles of population invariance. However, they employed the Rasc model to estimate the testlets rather than using testlet response theory. Therefore, this study uses the English test score data from the 2007 Taiwan Assessment of Student Achievement (TASA) as an example, investigating population invariance after equalization, and examining the effects of using different measurement models.