潛在類別分析與二階段群集分析分群效果之比較研究

Abstract

本研究目的在檢驗傳統群集分析中的二階段群集分析和潛在類別分析的分群能力差異，共分為兩個子研究。研究一透過模擬資料操弄外顯變項數量、外顯變項類型和樣本大小三個自變項，比較兩種分群方法的分群結果特徵和分群結果準確度，以作為選擇分群方法的參考。研究二以電腦化創造力測驗為工具，400位大專生為受試者，蒐集創造成品和創造歷程兩部分評分資料後，以兩種分群方法進行實際應用，比較分群結果差異並了解不同創造力表現類型的受試者在不同觀察變項上的表現差異與反應特徵。 研究一結果顯示，在外顯變項包含有類別變項時，潛在類別分析法的表現明顯優於二階段群集分析。另外隨著樣本人數和外顯變項個數增加，分類錯誤率亦隨之下降。研究二結果顯示，不論是創造成品或創造歷程，兩種分群方法的分群結果非常相似，最後決定的群集數均相同，且各群集的反應特徵亦相當接近。而透過創造成品的評分規準，可將受試者區分為「創造設計者」、「實用設計者」和「想像設計者」三種不同表現類型；透過創造歷程的評分規準，可將受試者區分為「創造操作能力優異」、「實用操作能力優異」和「材質運用能力不佳」三組。最後依據本研究結果可提出以下建議： 1、進行分群研究時，若外顯變項包含有類別變項，應選擇潛在類別分析法，較能確認分群準確性。此外若能增加樣本人數和外顯變項個數，也有助於降低分類錯誤率。 2、研究二結果可應用在創造力教學上，教師可根據不同創造力表現類型的學生分別設計相對應的教學活動，例如加強創造成品中實用性與創新性的連結，或是針對創造歷程中表現不足的操作能力加以訓練。

The purpose of this research is to compare the results of clustering using latent class analysis and two-step cluster analysis. The research is composed of two sub-researches, simulation and empirical study. In Study 1, the type of observed variables, the number of observed variables and sample size were be manipulated. And compare the features of clustering, the accuracy of clustering with two clustering methods. In Study 2, we apply two clustering methods on creativity performances and compare the results of clustering. Four hundreds university students were asked to complete the “computerized creativity test”. Their creating process were automatically recorded and scored by computer using seven criterions. The creating product was scored using nine criterions by two raters. The results of Study 1 indicate that latent class analysis performs better than two-step cluster, when the observed variables contain categorical variables. Besides, the number of observed variables and sample size has significant effect. The ANOVA result shows that the more observed variables and sample size, the less misclassification rate of both methods. The results of Study 2 show that the results of two clustering methods are very similar either in creating process or in creating product. In creating product, there are three types of creativity performances in the subjects which were named as “Creative designer”, “Practical designer”, and “Imaginable designer”. In creating process, there are three groups in the subjects too. They were named as “Excellence ability to creative operate”, “Excellence ability to practical operate”, and “Poor ability to use material”.

According to the results of this study, two recommendations were proposed: 1. It’s better to use latent class analysis, when the observed variables contain categorical variables. The results of latent class analysis have more correct decisions and less misclassification. In addition, to reduce the misclassification rate, researcher can add more observed variables and sample size. 2. Several pedagogical implications can be drawn from Study 2. Teacher can design different programs for different type of students, such as strengthening the link between innovative and practical of creating product, or provided more training for the poor performance of creating process.