Educational Research Journal



結構方程式之擬合優概念及常用指數之比較

1996.第11卷第1期(Vol. 11 No. 1).pp. 73–81
 

結構方程式之擬合優概念及常用指數之比較

侯傑泰、成子娟、鍾財文(Kit-Tai HAU, Zijuan CHENG, & Choi-Man CHUNG)

摘要

結構方程模式已成教育、社會及心理學等的重要數據分析技巧,本文主要是總結及引用數個新近發表有關擬合優指數的模擬數據研究結果。文章首先從近似及估計差距的概念,討論模式的適合性,繼而介紹數種常用擬合優指數,包括:交互效度確定、增值擬合、省儉及離中指數,除了描述各指數的設計理念及特徵外,亦討論一些評核指數的準則,如:獨立於樣本大小、懲罰覆雜模式、指數的信度等。在比較各指數時,文章亦推介RMSEA, RNI及NNFI作為一般常用的擬合優指數。

Abstract

Structural equation modeling has become an important statistical technique in educational, social, and psychological research. The present article summarizes and presents the results of a number of recently published Monte Carlo studies on goodness of fit indexes. First, the concept behind model fitting is discussed with reference to approximation and estimation discrepancies. Then, the philosophy and characteristics of several common categories of fit indexes are described, which include cross-validation, incremental, parsimony, and noncentrality indexes. Lastly, the criteria in evaluating an index are discussed and it is recommended that RMSEA, RNI and NNFI can be used for routine SEM analyses.