文章詳目資料

調查研究-方法與應用 TSSCI

  • 加入收藏
  • 下載文章
篇名 Mokken Scale 量表之應用: 以長期變遷的語言學習 和使用態度題組為例
卷期 39
並列篇名 Application of Mokken Scale: An Example of Attitudes toward Language Learning and Use with Longitudinal Data
作者 廖培珊江振東林定香
頁次 007-042
關鍵字 Mokken 量表葛特曼量表項目反應理論潛藏特質語言學習Mokken scaleGuttman scaleItem Response Theorylatent traitlanguage learningTSSCI
出刊日期 201804

中文摘要

調查問卷常使用計分量表來測度受訪者的潛藏特質,然而以常見 的葛特曼量表為例,題目間難易程度之嚴格假定,缺乏隨機性的概念 與測量誤差之考量等,使其應用受到限制。被視為項目反應理論之無 母數模型的Mokken 量表,具單一向度、局部獨立性、單調性等的特 質,並考量回答類別的反應機率,因此可用以檢測含數個題目的題組 是否可組成測量同一潛藏概念之量表。 本研究介紹Mokken 量表模型之基本假設與特性,並以「台灣社會變遷基本調查」2004 年至2014 年共三期的資料檔中有關語言學習 及使用態度之題組為例,利用Mokken 量表的條件特性來檢視量表組 成,並將估算出之量表分數與個人背景和語言使用之變項進行分析。 研究結果指出,以逐項條件檢驗的方式來看,包含三個題目的語言學 習及使用態度題組整體而言符合Mokken 量表之特性;邏輯斯迴歸分 析則可看出個人背景及家中使用語言與潛藏態度的關係,均會因調查 年度的不同而有差異,本研究並就Mokken 量表模型的特性與應用分 析結果提出討論。

英文摘要

Cumulative scales have been widely used in questionnaires to measure respondents’ latent traits, such as personality or attitudes toward certain events. Different scaling models have been developed to reveal the disparity among respondents, in which Guttman-type scales are commonly used. However, their assumption of a unique combination of response patterns and lack of stochastic implications of the scalagram indices have limited their application to empirical studies. The Mokken scale, which can also be seen as a non-parametric item nonresponse model (NIRT), has been developed to relax the rigid assumptions of the Guttman scale while maintaining its requirements of uni-dimensionality, local independence, and monotonicity. In addition, the Mokken scale takes into account probabilities of responses to question items, which enable the examination of whether these items are measures of the same latent construct. This study describes the basic assumptions and characteristics of the Mokken scale, and demonstrates its application using empirical data. Question items which measure the latent attitudes about language learning and use from 2004-2014 Taiwan Social Change Survey (TSCS) are used as an example to examine the scalability of the items and the scale that consists of those items. Furthermore, logistic regression of the latent trait using Mokken scale scores on demographic characteristics and a variable associated with language used at home are performed. Our results indicate that the association between the examined variables and the latent trait varies depending on the survey year. Discussion on the conditions of the Mokken scale model and findings based on empirical data are provided.