知識結構的測量：徑路搜尋法與概念構圖法的比較

Assessing Structural Knowledge: A Comparison of Pathfinder and Concept Mapping

徑路搜尋法（Pathfinder）是近年來應用十分廣泛的知識結構測量方法，概念構圖法（concept mapping）則是理論基礎十分完備的教學和評量策略。本研究嘗試將概念構圖法中「引出知識」和「表徵知識結構」的步驟、徑路搜尋法中的知識結構評量算則（C指數）以及Novak & Gowin（1984）的概念構圖計分法等加以結合，設計出概念構圖相似指數（CMC）及Novak-Gowin指數（N-G）。本研究以153名大學生為對象，以教育心理學的「學習理論」為材料，評估概念構圖法的兩個知識結構指數CMC及N-G，和徑路搜尋法的兩個知識結構指數PFC及PRX，對學習成就的預測效果、對不同成就學生的區別效果、與各指數的重疊與獨立性。結果有以下的發現：1.以徑路搜尋法求得的知識結構指數對學習成就的解釋量，PFC為36%，PRX為16%；以概念構圖法求得的兩個指數對學習成就的解釋量，CMC為36%，N-G為23%。兩種測量方法所提供的知識結構指數具有可接受的預測效度。2.應用四種知識結構指數，PFC、PRX、CMC及N-G來區別不同學習成就的學生時，PRX並沒有顯著的區別力。但PFC，CMC及N-G等三種指數可以有效地區別不同學習成就的學生，準確性達80%。3.藉由不同的引出知識、表徵知識結構和評量知識結構等步驟求出的PFC，CMC和N-G指數，彼此間具有中度的相關。以淨相關分析排除某一指數的影響效果後，其他兩種指數仍能維持與學習成就的顯著相關。可見以不同方法求得的知識結構指數彼此有其獨立的成份存在，未必為其他指數的成份所能涵括。

Pathfinder is a widely applied method for structural knowledge assessment. Concept mapping is a theory-based strategy of instruction and evaluation. In this study, we combined the 'knowledge elicitation' and 'knowledge representation' procedures of concept mapping, the knowledge-structure evaluation algorithms of Pathfinder, and the scoring rubrics of concept mapping proposed by Novak & Gowin (1984)to implement two indices for assessing structural knowledge--the concept mapping based closeness index (CMC) and the Novak-Gowin index(N-G).To explore the predictive and discriminant validities of CMC,N-G, the PFC and PRX indices obtained from Pathfinder, and the possible redundancy among them,153 college students' concept maps and similarity judgement data of 31 concepts from the 'learning theories', and their achievement scores of educational psychology, were collected. The results showed: 1. All the four indices were significantly correlated to the achievement score of educational psychology. The PFC, PRX, CMC, and N-G accounted for 36%, 16%, 36%, and 23% of the variances of the achievement score, respectively. 2. The results of discriminant analysis showed that the PFC, CMC, and N-G indices correctly discriminated 80% of the group memberships of the students of the high- or low achievement groups. Yet the PRX was excluded from the discriminant function. 3. The PFC, CMC, and N-G indices, though elicited and computed through different procedures, correlated significantly with each other. The result of partial correlation showed that when the influence of any index was controlled, the other indices were still correlated significantly with the achievement score. Therefore, though there might be redundant components among those indices, they did remain their own unique components.