HKIER

摘要

結構方程（SEM）已成多變量分析的重要技巧，本文檢視SEM較傳統分析優越之處，並討論如何用單指標變項、項目對（或小組）及單同源分子測量模式去建立潛伏變項；此外亦簡略探討偏態序級量表、最少樣本數目及吻合指數等課題。

Abstract

Structural equation modelling (SEM) has become a very important technique in multivariate analysis. The present article examines the advantages of SEM over traditional analytical methods. Construction of latent variables using single indicator, item pairs/parcels, and uni-factor congeneric factor model are discussed with concrete examples. Furthermore, issues related to skewed ordinal scales, minimum sample size, and choice of goodness of fit indexes are reviewed.