驗證性因素分析與概化部份計分模式之模擬比較

本研究主要在探討因素分析模式和概化部份計分分析模式對潛在特質估計的影響，並將因素分數、能力值視為潛在特質，進行比較。此外，亦並探討不同的鑑別度、資料合併需求與模式的交互影響，研究者透過比較各模式相關、RMSD估計的精確，及變異數分析，得到如下的結論。1、在嚴格控制均勻的鑑別度、資料合併需求下，若只求相關值，概化部份計分並無較因素分析估計精準。2、透過變異數分析，不論模式、長度、鑑別度、資料合併需求皆達顯著。3、由因素分析所得的資料隨著因素負荷量、鑑別度、長度之上升與資料合併需求之下降，相關值規律上升，RMSD規律下降。4、本研究仍未能證明因素分析優於概化部份計分模式，因為在一份真實的測驗，鑑別度、資料合併需求並不會完全均等。

關鍵字

因素分數；因素分析；概化部份計分模式

並列摘要

The main objective in the study is to analyze whether the use of different analytic model have an impact on latent trait estimation for confirmatory factor analysis and generalization partial credit model (GPCM). In the study, ability and factor score are regarded as latent trait. In addition, the study also examines whether test length discrimination and the need for collapsing interact with the models. For this reason, ANOVA is used. The cutoff point of the difference of RMSD and correlation between true latent trait and estimation which is obtained from stimulation is also used. By these ways, some results are inferred. First, under the same discrimination and need for collapsing, GPCM never surpasses factor analysis. Second, by ANOVA, it is significant regardless of any discrimination need for collapsing length and model. Third, the correlation grows and RMSD decline with the increase of discrimination, length and factor loading and decrease of need for collapsing by the data which is analyzed by factor analysis but not by GPCM. Lastly, in this research it is not proven that factor analysis excels GPCM, because in the true test, the discrimination and need for collapsing is not the same with each other.