中介作用的意涵及其資料分析策略之探究

The Conceptualization and Data Analytical

背景：中介作用的概念長久來即見於心理學的論述中，但實證性地探究中介作用則直至1980 年代，Baron 與Kenny（1986）、James 與Brett（1984）及Judd 與Kenny（1981）等幾位學者構建資料分析的策略後才開始普及。目前中介作用的探究已廣見於社會科學、健康學、心理學、教育及社會學的研究中（Wu & Zumbo, 2008）。然遺憾的是，中介作用的因果本質常被忽視或誤解，因而導致許多誤用或錯誤的解釋（Frazier, Tix, &Baron, 2004；MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002）。目的：本文的目的在於釐清中介作用的基本概念，並探究驗證中介作用的資料分析策略，期能有助於研究者進行中介作用的探究。討論焦點：除釐清中介作用的基本概念外，本文著重於討論Kenny 的因果步驟、係數乘積考驗法中的Sobel 考驗法及拔鞋法（bootstrapping）等三種檢證中介作用效果的方法，並比較其優缺點。結論：檢驗中介作用的方法均有其使用上的限制，進行分析時必須審慎。檢證中介作用效果時，可考慮採行多重的檢驗方法。同時並用數種檢驗方法，其所獲致的結果或許更為真確。

Background: The methodology of mediation is commonly used in social science, health,psychological, educational, and sociological research (Wu & Zumbo, 2008). Unfortunately,mediation effects have been widely misunderstood and misused as data analytical tools and statistical hypothesis testing with little theoretical and methodological considerations. (Frazier,Tix, & Baron, 2004; MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002).Aims: The current paper aims to clarify the concept of mediation and investigate data analytical strategies for testing mediation effects. Hopefully, the results could assist researchers in designing and conducting their studies when mediation effects are examined.Discussion focus: In addition to clarifying the concept of mediation, this paper compares three analytical approaches for testing mediation effects, namely, Kenny's causal steps analysis, Sobel test, and bootstrap procedure. Besides data analyzing procedures, strengths and weaknesses of these methods are discussed. Conclusion: These data analytical methods all have their limitations. Researchers should use them with caution. An integrated strategy is suggested when mediation effects are tested. By examining the testing results obtained from each approach simultaneously, it is more likely to gain a valid result.