利用分類與迴歸樹探討中學生學習成就的相關因素
No Thumbnail Available
Date
2009
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
本研究主要的目的在透過分類與迴歸樹（CART）分析方法，探討中學生在不同學習階段的學習成就情形及其相關因素，比較不同學習階段中學生之學習成就相關因素的差異，並進一步建立不同學習階段學習成就高低的分類預測模型。本研究是以臺灣教育長期追蹤資料庫（TEPS）為資料來源，採用第一波國中樣本和第三波高中/高職/五專追蹤樣本的學生和家長問卷，樣本數為3022人。經過CART分析之後，本研究有以下三點發現：
1. 中學生於不同學習階段的學習成就有顯著差異，高中/高職/五專學習階段的學習成就表現高於國中學習階段。
2. 中學生於不同學習階段的學習成就相關因素是有差異的。
（1）國中階段的CART分類模型包括11個變項，涵括了個人、家庭和社會網絡三個因素。其整體分類正確率達68.9%，經交互驗證法評估後略降為64.7%，對於高學習成就的分類正確率達79.6%，高於低學習成就的56.1%。
（2）高中/高職/五專階段的CART分類模型，包括課程類別和學校公私立別兩個變項，涵括了個人和學校兩個因素。其整體分類正確率達77.2%，經交互驗證法評估後仍達77.2%，對於高學習成就的分類正確率達80.1%，高於低學習成就的73.4%。
3. 中學生於不同學習階段，用來區分學習成就高低的因素有差異。
（1）就國中階段而言，如要對高學習成就進行預測，其主要解釋路徑經過「能力期望」、「父親教育程度」和「電腦使用時間」三個變項；次要解釋路徑經過「能力期望」、「父親教育程度」、「電腦使用時間」和「學校主動與父母聯繫程度」四個變項。另一方面，如要對低學習成就進行預測，其主要解釋路徑經過「能力期望」和「自我期望」兩個變項；次要解釋路徑經過「能力期望」、「自我期望」、「父母教育期望」和「父親職業」等四個變項。
（2）就高中/高職/五專階段而言，高低兩個不同程度的學習成就，其主要解釋路徑和次要解釋路徑所經過的變項是一樣的，但分類條件不同。如要對高學習成就進行預測，其主要解釋路徑經過「課程類別」（分類條件為普通學程自然組）一個變項；次要解釋路
徑經過「課程類別」（分類條件為普通學程非自然組、普通學程自然組、綜合學程：學術導向）和「學校公私立別」（分類條件為公立）兩個變項。另一方面，如要對低學習成就進行預測，其主要解釋路徑經過「課程類別」（分類條件為商業、綜合學程非學術導向、工業類、理工、文商、醫、藝術類、農業類、家事類、普通科、海事水產類）一個變項；次要解釋路徑經過「課程類別」（分類條件為普通學程非自然組、綜合學程學術導向）和「學校公私立別」（分類條件為私立）兩個變項。
最後，根據本研究的發現提出相關討論及未來研究的建議。
The purpose of this study was to, through Classification and Regression Trees (CART) analysis, investigate and compare leaning achievements and relevant factors among high school students in different learning stages, and then establish a classification model that can predict different levels of learning achievement in different learning stages of high school students. The data of this longitudinal study was collected from the database of Taiwan Education Panel Survey, adopting waveⅠand wave Ⅲ questionnaires filled out by one selected group of students and their parents. The total sample contained 3022 students who had joined in junior high school in waveⅠproject and followed up as they were senior high school/vocational high school/junior college students in wave Ⅲ. Data were analyzed by CART. The three major findings of this study were shown as follows. 1. The leaning achievements of high school students in different learning stages were significantly different. The learning achievements of senior high school/vocational high school/junior college students were better than that of junior high school students. 2. The learning achievement related factors in different learning stages of high school students were different. (1) The CART classification model for junior high school stage included three factors that concern individual, family, and social network, within which 11 variables were revealed. The total accuracy of this classification model was 68.9%, while decreasing to 64.7% under the subsequent cross-validation evaluation. For high learning achievement, classification accuracy was 79.6%, relatively better than that for low learning achievement, which was 56.1%. (2) The CART classification model for senior high school/vocational high school/junior college stage included two factors that concern individual and school, which were the two variables, course type and public or private school. The total accuracy of classification model was 77.2% and remained the same after cross-validation evaluation. For high learning achievement, classification accuracy was 80.1%, to some degree better than that for low learning achievement, which was 73.4%. 3. The factors which were used to discriminate different levels of learning achievement in different learning stages of high school students were distinguishable. (1) In junior high school stage, the main explaining path of predicting high learning achievement passed through three variables, including ability expectancy, father’s education degree, and the time spent on computer. The secondary explaining path of predicting high learning achievement passed through four variables, including ability expectancy, father’s education degree, the time spent on computer, and school’s active contact frequency. Besides high learning achievement, we could predict low learning achievement with its main explaining path passing through two variables that were ability expectancy and self-expectancy, or with the secondary explaining path through four variables that were ability expectancy, self-expectancy, parents’ expectation and father’s occupation. (2) In senior high school/vocational high school/junior college stage, the main and secondary explaining path of predicting both high and low learning achievement passed through same variables, namely, course type and public or private school, but the two categories hold different classification conditions. For high learning achievement, the main explaining path passed through course type and its classification condition was majoring in natural sciences in general courses, and its secondary explaining path passed through course type (whose classification condition was majoring in academic-oriented courses such as non-natural science, natural science, and comprehensive course), and through public and private school, with public school as the classification condition. On the other hand, for low learning achievement, the main explaining path passed through course type and its classification condition was majoring in business, non-academic-oriented comprehensive courses, industry, technology, literature, medicine, art, agriculture, household management, general courses, and maritime affairs and aquatic products. The secondary explaining path for low learning achievement passed through course type, with the classification condition as majoring in general courses and academic-oriented comprehensive courses, and through public or private school, while the classification condition is private school. Finally, these findings were discussed and the direction of future research and application were suggested.
The purpose of this study was to, through Classification and Regression Trees (CART) analysis, investigate and compare leaning achievements and relevant factors among high school students in different learning stages, and then establish a classification model that can predict different levels of learning achievement in different learning stages of high school students. The data of this longitudinal study was collected from the database of Taiwan Education Panel Survey, adopting waveⅠand wave Ⅲ questionnaires filled out by one selected group of students and their parents. The total sample contained 3022 students who had joined in junior high school in waveⅠproject and followed up as they were senior high school/vocational high school/junior college students in wave Ⅲ. Data were analyzed by CART. The three major findings of this study were shown as follows. 1. The leaning achievements of high school students in different learning stages were significantly different. The learning achievements of senior high school/vocational high school/junior college students were better than that of junior high school students. 2. The learning achievement related factors in different learning stages of high school students were different. (1) The CART classification model for junior high school stage included three factors that concern individual, family, and social network, within which 11 variables were revealed. The total accuracy of this classification model was 68.9%, while decreasing to 64.7% under the subsequent cross-validation evaluation. For high learning achievement, classification accuracy was 79.6%, relatively better than that for low learning achievement, which was 56.1%. (2) The CART classification model for senior high school/vocational high school/junior college stage included two factors that concern individual and school, which were the two variables, course type and public or private school. The total accuracy of classification model was 77.2% and remained the same after cross-validation evaluation. For high learning achievement, classification accuracy was 80.1%, to some degree better than that for low learning achievement, which was 73.4%. 3. The factors which were used to discriminate different levels of learning achievement in different learning stages of high school students were distinguishable. (1) In junior high school stage, the main explaining path of predicting high learning achievement passed through three variables, including ability expectancy, father’s education degree, and the time spent on computer. The secondary explaining path of predicting high learning achievement passed through four variables, including ability expectancy, father’s education degree, the time spent on computer, and school’s active contact frequency. Besides high learning achievement, we could predict low learning achievement with its main explaining path passing through two variables that were ability expectancy and self-expectancy, or with the secondary explaining path through four variables that were ability expectancy, self-expectancy, parents’ expectation and father’s occupation. (2) In senior high school/vocational high school/junior college stage, the main and secondary explaining path of predicting both high and low learning achievement passed through same variables, namely, course type and public or private school, but the two categories hold different classification conditions. For high learning achievement, the main explaining path passed through course type and its classification condition was majoring in natural sciences in general courses, and its secondary explaining path passed through course type (whose classification condition was majoring in academic-oriented courses such as non-natural science, natural science, and comprehensive course), and through public and private school, with public school as the classification condition. On the other hand, for low learning achievement, the main explaining path passed through course type and its classification condition was majoring in business, non-academic-oriented comprehensive courses, industry, technology, literature, medicine, art, agriculture, household management, general courses, and maritime affairs and aquatic products. The secondary explaining path for low learning achievement passed through course type, with the classification condition as majoring in general courses and academic-oriented comprehensive courses, and through public or private school, while the classification condition is private school. Finally, these findings were discussed and the direction of future research and application were suggested.
Description
Keywords
學習成就, 分類與迴歸樹, 臺灣教育長期追蹤資料庫, learning achievement, Classification and Regression Trees (CART), Taiwan Education Panel Survey (TEPS)