次級量尺分數估計法於大型教育測驗之模擬研究

The Subscale Scores Estimation for Large-Scale Assessments

近幾年，次級量尺分數的估計方法與應用開始被重視，例如：國內外大型測驗（TIMSS、PISA、NAEP、TASA）的分數報告，均呈現不同能力向度之次級量尺分數。然而，雖然國外已有學者針對次級量尺分數之研究進行探討，但是國內部分目前尚無相關研究，且並沒有研究比較這些方法使用於等化測驗設計。因此，本研究主要以模擬實驗方式探討不同次級量尺分數計算方法於不同測驗情境中，對於單一測驗設計與等化測驗設計分數之估計效果。此外，本研究亦提出新的次級量尺分數計算方法，以比較不同次級量尺分數計算方法之差異。研究結果發現，本研究提出之新的次級量尺計算方法，於不同測驗情境中具有較佳之估計精準度。

The purpose of this paper is to explore subscale scores estimation in two testing design situations， single testing design and equating testing design. Additionally， two new methods to estimate subscale scores are presented in this paper.Using simulation data， this study investigates the accuracy of subscale scores estimation for different methods of estimating subscale scores. In single testing design， factors taken into consideration include the following: correlation between subscales， samplesizes， ratio of CR/MC items， numbers of subscales， and test length. In equating testing design， factors taken into consideration include the following: correlation between subscales， sample sizes， collocation of anchor items， and equating methods.The results show that:1. New methods of estimating subscale scores are better than other methods.2. The estimation error decreases as correlation between subscales increases; however，the sample sizes don’t impact the estimation error.3. In single testing design， the estimation error decrease as ratio of CR/MC itemsincrease and the estimation error decrease as test length increase.4. In equating testing design， the collocation of anchor items do not impact the estimation error and the concurrent calibration method based on item response theory has higher accuracy than equating calibration based on classical test theory.