以統合研究建立相互教學有效性的在地證據：研究方法特徵影響評估

Local Evidence-based Research Synthesis: How Methodological Features Moderate the Effects of Reciprocal Instruction

檢視過去統合研究，發現學者多關注文獻出版類型可能導致偏誤，鮮少針對研究方法特徵於統合結果影響進行評估與討論。據此，本文目的在探究研究方法特徵（實驗研究設計、評量工具與教學者來源）對於相互教學在促進不同學習階段學生，中文閱讀理解統合成效的影響。研究結果顯示，相互教學對於學生中文閱讀理解平均效果量為近中等程度教學成效，此外，進一步評估研究方法特徵影響時，發現當納入實驗研究設計因子，相互教學對參與者為國小高年級（β＝-0.2736，pβ＝0.0048）、或國中以上學生（β＝-0.2427，pβ＝0.0481）之教學效果皆顯著低於參與者為國小中年級學生。然而，評量工具來源、教學者是否為研究者則未產生顯著影響。最後，本文針對分析結果，提出未來實務與研究建議。

Systematic review and meta-analysis, one of the effective and widespread techniques in forming research-based and holistic evidences, was widely accepted by researchers. With the evolution of this technique, the approach in assessing quality of primary research has become an important topic in this area. However, except publication bias, research has seldom been conducted to investigate biases deriving from methodological features under the context of research synthesis. The purpose of this paper is to evaluate the moderating effects concerning experimental design, measurement tools, and instructor-as-researcher, while investigation the effects of reciprocal instruction on Chinese reading comprehension across students’ grades. Systematic literature review processes were implemented to search, identify, and collect relevant studies. After few keywords were made in searching Taiwan literature database, a total of 2,437 articles were first retrieved, and followed a two-step literature screening process according to inclusion and exclusion rules, 43 articles were formally adopted. Random effect model was adopted while meta-analyzing findings of the included studies, an overall average effect size of +0.4756 and +0.3969 were obtained separately for immediate and follow-up effect. As to the analysis of moderator effects, the findings showed a significant lower average effect size separately for students in fifth-to-sixth grade, and above middle school ages in comparison to students in third-to-fourth grade while using multiple meta-regression with controlling the impact of experimental design. However, no significant moderating effects were found in measurement tools, and instructor-as-researcher. Finally, suggestions and implications were raised for further research.