國家教育研究院

　　NAEP、TIMSS和PISA等國際大型測驗，學生的成就資料是以「可能值」的資料型態提供給次級資料的分析者。可能值方法是以潛在迴歸模式，加入學生背景變項計算後驗分佈，並抽取可能值，以利於次級資料分析者使用。目前可能值方法的相關研究主要著重於可能值估計方法的改善、可能值方法與點估計方法間比較、可能值方法估計不同計分模式之比較及探討背景變項的相關程度對於可能值方法之影響，然這些研究主要著重於單向度能力值估計方法的改善與比較，且尚未有文獻探討不同的等化設計，對於可能值方法的參數估計之影響。隨著測驗的複雜化，目前大型測驗多為多向度取向且採用不同的等化設計，因此本研究擬採用模擬資料以及實徵資料進行分析與探討，不同相關程度的背景變項，以及不同等化設計，對於單向度可能值方法以及多向度可能值方法的影響，並提出較適當的估計方法以及實驗設計，提供給國內大型測驗未來實行之參考依據。

　　Large-scale assessment such as NAEP, TIMSS and PISA were concert about the population and the parameters of population. In those assessment student’s achievement data sets provided for secondary analysts contain so-called plausible values. Plausible values are random draws from a student’s posterior distribution. Plausible values can be used to estimate population characteristics, and it can lead to unbiased estimates of population parameters. Most of the references were focus on the recovery of population parameters by comparing point estimates and plausible values. Those research only concert about the unidimensional estimation method, but as the assessment progressing the test become complicate and multidimensional. Besides, this study will discuss the different correlation between background variables and latent variable. Further, those Large-scale assessment adopt some equate design, but there were no reference to point out the effects of the parameters estimator. Therefore, the purpose of this study were using simulation and empirical data to explore that different correlation and equating design effects unidimensional and multidimensional plausible values methods. Provides appropriate estimation methods and manipulate design to the Large-scale assessment.