以徑路搜尋法表徵內隱與外顯知識結構的效度探討
No Thumbnail Available
Date
2005
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
本研究運用四種概念接近性測量法：促發法、測驗法、評定法、Q分類法，以421位國小六年級學生為測量對象，以自然科為測量內容，由四位專家教師從教材中挑選出14個重要概念，利用四種測量法測量兩兩概念相似性，並將測量結果分別進行徑路搜尋法分析，主要目的在於探討：(一)促發法的測量結果是否具有建構效度；(二)各種相似性指數及測量法對學習成就的解釋力有無差異；(三)由各種測量法產生的知識結構是否具有不同的特點；(四)MTMM與MDMM基準模式的理論假設能否獲得支持。基於以上目的，本研究提出新的rPF指數(r of Pathfinder)及MDMM（multidimension-multimethod）模式，進行實徵檢驗。研究工具包括概念相似性評定問卷、知識結構測驗、學習成就測驗、SuperLab實驗程式、概念相似性分類卡片等五項。研究結果如下：
一、就促發法的建構效度而言
以促發法的測量結果，進行徑路搜尋法分析後產生的相似性指數，若作為「知識結構」的觀察指標，能夠符合Reber（1993）提出的「年齡獨立性」、「低變異性」、「智力獨立性」等內隱記憶系統的假設；亦即以徑路搜尋法之語意網路模式來表徵促發法的測量結果，所形成的「內隱知識結構」能夠符合相關的理論假定。而促發法的測量結果也具有適當的內部一致性信度與輻合效度。
二、就相似性指數和測量法的效標關聯效度而言
(一) rPF指數若作為潛在變項「知識結構」的指標對學習成就進行迴歸，在評定法和Q分類法中具有顯著的解釋力。若直接對學習成就進行迴歸，rPF指數在測驗法中具有顯著的解釋力；rPF指數在測驗法中也能顯著增加不同於其它三項指數的解釋量。
(二) 若直接對學習成就進行迴歸，以PRX指數的解釋力最強，其次為PFC指數。
(三) 以相似性指數作為潛在變項「知識結構」的指標對潛在變項「學習成就」進行迴歸，以評定法對學習成就的解釋量最高，其次為測驗法。
三、就各種測量法產生的知識結構圖的比較而言
(一) 就參照結構的比較而言，其特點包括了：
1. 連結數不同：以促發法產生的連結數最少，測驗法產生的連結數最多。
2. 可解釋性不同：促發法和測驗法產生的參照結構，與教材差異較大，而且有部份的連結不容易以教材內容直接加以解釋。評定法和Q分類法產生的參照結構，與教材較相似，所有節點之間都具有符合教材內容的邏輯關係。
3. 相似性不同：促發法的參照結構與其他測量法的參照結構之差異較大，而評定法和Q分類法的參照結構兩者之間有較大的相似性。
(二) 若以兩位高成就的受試者之知識結構為代表，與參照結構的分析結果相比較，可發現在「連結數」、「可解釋性」、「相似性」等三方面，參照結構的分析結果多可以類推到高成就受試者的知識結構之上。
(三) 此外，受試者的知識結構之核心概念的一致性也因測量法而有不同：其中以測驗法較大，其參照結構和受試者的知識結構都包括了相同的核心概念；但其他三種測量法則未發現此現象。
四、就MTMM和MDMM之模式考驗而言
(一) 本研究結果支持MTMM競爭模式2的假定：「內隱知識結構和外顯知識結構是兩個獨立無關的特質」。
(二) 本研究結果支持MDMM基準模式的假定：「節點內容（PFC指數）、節點圖解距離（GTD指數）、節點原始距離（PRX指數）等三個向度之間有相關」。
最後，研究者根據研究結果，提出未來應用徑路搜尋法進行相關研究、或者探討本研究提出的新潛在變項「內隱知識結構」等在教學與研究上的建議。
This study used four measuring methods, Priming, Test, Rating, and Q-sort, to examine the concept similarities on science from 421 elementary school students. 14 important concepts were chosen by 4 expert teachers to examine their similarities, and these results were analyzed by Pathfinder Analysis. The purposes of this study was to probe: (1) whether the results measured by Priming has the construct validity; (2) whether the explanatory power of each similarity index and measuring method have differences on learning achievement; (3) whether the knowledge structure formed by each measuring method has different characteristics; (4) whether the assumptions of MTMM (multitrait-multimethod) and MDMM (multidimension-multimethod) models are supported. Based on these purposes, new rPF index and MDMM model were suggested to proceed a field study by concept similarities questionnaire, knowledge structure test, learning achievement test, SuperLab computer program, and concept similarities sort-card. The results of this study were found as follows: 1. About the construct validity of Priming method Using the similarity indices conducted by Pathfinder Analysis on the measuring results of Priming method to represent knowledge structure can fit the assumptions of implicit memory systems on Age independence, Low variability, and IQ independence. Namely, the semantic network models in Pathfinder Analysis were adopted to represent the results measured by Priming method. Moreover, the results measured by Priming method meet the internal consistency reliability and the convergent validity as well. 2. About the criterion-related validity of similarity indices and four measuring methods (1) Using rPF index as an indicator of the latent variable “knowledge structure” to proceed regression analysis on learning achievement has significant explanatory power in Rating method and Q-sort method. Besides, proceeding regression analysis on learning achievement makes rPF index has significant explanatory power in Test method. Moreover, rPF index in Test method can also increase the explanatory power different from the ones by the other three indices: PFC, GTD, and PRX. (2) PRX index has the highest explanatory power when proceeding regression analysis on learning achievement, and PFC index is the next. (3) When using four similarity indices as indicators of the latent variable “knowledge structure” to proceed regression analysis on the latent variable “learning achievement”, Rating method has the highest explanatory power, and Test method is the next. 3. About the comparisons of knowledge structures formed by different measuring methods (1) The characteristics of comparing with each referent structure: a. Different links: Priming method has the least links (13), and the Test method has the most (17). b. Different explanatory power: referent structures formed by Priming and Test methods were more different to teaching materials, and some links were hard to interpret by teaching content; referent structures formed by Rating and Q-sort methods were more similar to teaching materials, and all links meet the logic of teaching content. c. Different similarities: referent structure formed by Priming method was more different from the ones formed by other measuring methods; referent structure formed by Rating method was more similar to the one formed by Q-sort. (2) Comparing the knowledge structures of two high-achieving subjects with the results of referent structures in “links”, “explanatory power”, and “similarities”, most results of referent structures can generalize to the knowledge structures of high-achieving subjects. (3) Furthermore, the consistencies on central concepts among knowledge structures of these two subjects are distinctive due to different measuring methods. Test method has more consistencies: its referent structure and knowledge structure of subjects show the same central concepts. This result was not found in other three methods 4. About the test of MTMM and MDMM models (1) This study accepted the assumptions of competing model 2 in MTMM: implicit and explicit knowledge structures were two independent traits. (2) This study accepted the assumptions of baseline model in MDMM: PFC, GTD, and PRX are relevant. Suggestions for future researches conducted by Pathfinder Analysis or probe into the new latent variable “implicit knowledge structure” found in this study were provided in the end.
This study used four measuring methods, Priming, Test, Rating, and Q-sort, to examine the concept similarities on science from 421 elementary school students. 14 important concepts were chosen by 4 expert teachers to examine their similarities, and these results were analyzed by Pathfinder Analysis. The purposes of this study was to probe: (1) whether the results measured by Priming has the construct validity; (2) whether the explanatory power of each similarity index and measuring method have differences on learning achievement; (3) whether the knowledge structure formed by each measuring method has different characteristics; (4) whether the assumptions of MTMM (multitrait-multimethod) and MDMM (multidimension-multimethod) models are supported. Based on these purposes, new rPF index and MDMM model were suggested to proceed a field study by concept similarities questionnaire, knowledge structure test, learning achievement test, SuperLab computer program, and concept similarities sort-card. The results of this study were found as follows: 1. About the construct validity of Priming method Using the similarity indices conducted by Pathfinder Analysis on the measuring results of Priming method to represent knowledge structure can fit the assumptions of implicit memory systems on Age independence, Low variability, and IQ independence. Namely, the semantic network models in Pathfinder Analysis were adopted to represent the results measured by Priming method. Moreover, the results measured by Priming method meet the internal consistency reliability and the convergent validity as well. 2. About the criterion-related validity of similarity indices and four measuring methods (1) Using rPF index as an indicator of the latent variable “knowledge structure” to proceed regression analysis on learning achievement has significant explanatory power in Rating method and Q-sort method. Besides, proceeding regression analysis on learning achievement makes rPF index has significant explanatory power in Test method. Moreover, rPF index in Test method can also increase the explanatory power different from the ones by the other three indices: PFC, GTD, and PRX. (2) PRX index has the highest explanatory power when proceeding regression analysis on learning achievement, and PFC index is the next. (3) When using four similarity indices as indicators of the latent variable “knowledge structure” to proceed regression analysis on the latent variable “learning achievement”, Rating method has the highest explanatory power, and Test method is the next. 3. About the comparisons of knowledge structures formed by different measuring methods (1) The characteristics of comparing with each referent structure: a. Different links: Priming method has the least links (13), and the Test method has the most (17). b. Different explanatory power: referent structures formed by Priming and Test methods were more different to teaching materials, and some links were hard to interpret by teaching content; referent structures formed by Rating and Q-sort methods were more similar to teaching materials, and all links meet the logic of teaching content. c. Different similarities: referent structure formed by Priming method was more different from the ones formed by other measuring methods; referent structure formed by Rating method was more similar to the one formed by Q-sort. (2) Comparing the knowledge structures of two high-achieving subjects with the results of referent structures in “links”, “explanatory power”, and “similarities”, most results of referent structures can generalize to the knowledge structures of high-achieving subjects. (3) Furthermore, the consistencies on central concepts among knowledge structures of these two subjects are distinctive due to different measuring methods. Test method has more consistencies: its referent structure and knowledge structure of subjects show the same central concepts. This result was not found in other three methods 4. About the test of MTMM and MDMM models (1) This study accepted the assumptions of competing model 2 in MTMM: implicit and explicit knowledge structures were two independent traits. (2) This study accepted the assumptions of baseline model in MDMM: PFC, GTD, and PRX are relevant. Suggestions for future researches conducted by Pathfinder Analysis or probe into the new latent variable “implicit knowledge structure” found in this study were provided in the end.
Description
Keywords
徑路搜尋, 內隱知識, 外顯知識, 知識結構, 促發