中介效果的時序設計與因果推論

Sequence Design and Causal Inference of Mediation Effect

洪兆祥(Hung Chao-Hsiang)

指導教授 ： 余民寧

國立政治大學/教育學院/教育學系/博士(2018年)

https://doi.org/10.6814/DIS.NCCU.EDU.020.2018.F02

摘要

中介模型的本質在於解釋研究變數之間的因果傳遞機制，乃是當代社會科學領域中最常被使用的研究架構。然而大多數研究使用的橫斷中介設計，缺乏時間因素將難以進行因果推論的判斷，嚴重危害研究結果的正確性。本研究旨在探討中介模型如何進行完備的因果推論。其研究目的之一：探究中介模型的時序設計與因果推論的關係。其研究目的之二；探討現存的中介模型在因果推論上的完備條件。其研究目的之三：提出增強因果推論完備條件的可能策略為何。
本研究共有兩個子研究：研究一以443位國三學生為對象，採用五次模擬考成績為效標變數，並以「數學焦慮量表」、「數學學習動機量表」為工具，進行為期一年的追蹤測量。經描述統計分析、相關分析、路徑分析與中介檢定，結果顯示：
1. 在橫斷中介模型當中，數學焦慮為原因的因果推論較具完備條件。
2. 在序列中介模型當中，數學焦慮為原因的因果推論較具完備條件。
3. 在縱貫中介模型當中，數學焦慮為原因的因果推論，與數學學習動機為原因的因果推論，兩者的完備條件相當。
研究二以台灣青少年成長計畫資料庫當中的1529位國一學生為對象，分別選取親子互動、自尊與憂鬱做為研究變數，在八年之間時進行三次追蹤測量。經描述統計分析、相關分析、路徑分析與中介檢定，結果顯示：
1. 在橫斷中介當中，自尊的中介效果達顯著水準。
2. 在序列中介當中，自尊的中介效果達顯著水準。
3. 在縱貫中介當中，自尊的中介效果未達顯著水準。
本研究根據資料分析結果進行討論，並對未來研究提出建議。

關鍵字

中介模型 ； 時序設計 ； 因果推論

並列摘要

The purpose of mediation model is to explain the causal transfer mechanism among variables and it is a commonly used research framework in social science. However, most of the studies adapted cross sectional mediation analysis, which lacked sufficient time in making causal inferences. Such a practice may lead to a severe inaccuracy in data interpretation.
Therefore, this study aims to investigate how mediation models can achieve comprehensive causal inferences. The first research purpose is to examine the relationship between sequence design and causal inference of mediation models. The second research purpose is to investigate the comprehensiveness of the causal inferences in the existing mediation models. The third research purpose is to discuss possible strategies that can be used to enhance causal inference.
There are two research conducted in this study. For the first research, the participants were 443 ninth graders underwent a one year panel measurement. Five mock exam grades were used as the criterion variable, and the anxiety of mathematic exam and the motivation of learning mathematic were the research variables. With descriptive analysis, correlational analysis, path analysis, and mediation test, the results are as follow:
1. In the cross-sectional mediation model, it is more adequate to assume the anxiety of mathematic exam as the reason from causal inference. 2. In the sequential mediation mode, it is more adequate to assume the anxiety of mathematic exam as the reason from causal inference. 3. In the longitudinal mediation model, the anxiety of mathematic exam and the motivation of learning mathematic as the reason from causal inference are equally adequate.
　　The second research included 1529 seventh graders from TYP databased. Parent-child interaction, self-esteem and depression were used as the variables, and three panel measurements were taken over eight years. With descriptive analysis, correlational analysis, path analysis, and mediation analysis, the results are as follow:
1. In the cross-sectional mediation model, the mediation effect of self-esteem was significant. 2. In the sequential mediation mode, the mediation effect of self-esteem was significant. 3. In the longitudinal mediation model, the mediation effect of self-esteem was not significant.
The results and findings are discussed and implications for future research are also suggested.

並列關鍵字

Mediation effect ； Sequence design ； Causal inference

參考文獻

王金香（2010）。焦慮與動機影響數學學習之縱貫研究（未出版之博士論文）。國立政治大學，臺北市。

伊慶春(2012a)。台灣青少年成長歷程研究：國一樣本(J1)第四波，青少年問卷 (公共版)(C00179_3)【原始數據】取自中央研究院人文社會科學研究中心調查研究專題中心學術調查研究資料庫。

伊慶春(2012b)。台灣青少年成長歷程研究：國一樣本(J1)第八波，青少年問卷 (公共版)(C00272_2)【原始數據】取自中央研究院人文社會科學研究中心調查研究專題中心學術調查研究資料庫。

伊慶春(2014)。台灣青少年成長歷程研究：國一樣本(J1)第一波，青少年問卷 (公共版)(C00176_4)【原始數據】取自中央研究院人文社會科學研究中心調查研究專題中心學術調查研究資料庫。

余民寧(2013)。縱貫性資料分析—LGM的應用。臺北市：心理。