當PLS遇上SEM：議題與對話

When PLS Meets with SEM: Issues and Dialogues

邱皓政(Haw-Jeng Chiou)

《αβγ量化研究學刊》 Volume 3 Issue 1 (2011/06) Pp. 20-53

Abstracts

在社會科學領域，結合因素分析與路徑分析的結構方程模式(SEM)，自七十年代以來即為重要的量化方法典範，在教育、心理、管理乃至於實務領域，對於SEM的應用均不陌生。但另一方面，以迴歸預測分析為目的的淨相關最小平方法(PLS)近年來則受到行銷、資管、作業研究等管理領域學者的重視，最近PLS在工具發展與議題擴散則有加速的趨勢。有趣的是，擅長於這兩個方法典範的研究者未必對性質相似的另種方法學有所瞭解，導致兩個典範各行其是缺乏對話。本文的主要目的，是在比較SEM與PLS兩種分析典範在統計原理與應用實務上的差異，並以國家教育研究院的TASA資料庫中的實徵資料進行PLS、SEM、典型相關、主成份分析與探索性分析的比較示範。希望藉由當代量化研究社群對於這兩種方法學的一些討論，以及實徵數據的演示說明，勾勒出這兩種量化典範未來發展的趨勢。

Keywords

偏最小平方法 ； 結構方程模式 ； 形成性模式 ； 反映性模式

Parallel abstracts

Structural equation modeling (SEM) integrates the factor analysis with path analysis into a single framework which becomes an important statistical paradigm in social science since 70s. Researchers of education and psychology mostly familiar with SEM. On the other side, partial least square (PLS) is also widely used in some fileds such as marketing, information management and operational management in the recent decay. Interestingly, both paradigms share the similar function, however, little known to each other for both camps. The purpose of this paper is to compare the statistical principles and applications of the SEM and PLS. An empirical example using partial data from TASA established by National Institute of Education of Taiwan is demonstrated the similarity and difference among SEM and PLS as well as canonical correlation analysis, principal component analysis, and exploratory factor analysis. The trend of future development and potential coporation of both paradigms are discussed in the final section of this paper.

Parallel keywords

structural equation modeling ； partial least square ； formative model ； reflective model