Mokken量尺分析的實務運用：以「英語閱讀能力測驗」與「英語口說情緒反應量表」為例

Mokken Scale Analysis for Educational and Psychological Testing: Two Applications

Mokken量尺分析之目的在建立符合非參數單調同質模型或雙重單調同質模型的單一潛在量尺，其程序包含篩選試題的規準與逐一檢核試題符合模型假定的方法。本研究旨在應用此非參數試題反應理論，分別對2,270位大一新生的「英語閱讀能力測驗」與168位國中學生的「英語口說情緒反應量表」進行分析。結果在10題二元計分的閱讀能力測驗分析發現，共有5題符合雙重單調模型假設；在20題多元計分的口說情緒反應題目檢驗發現，僅有2題違背單調同質性模型。兩份資料的題目反應都符應單一潛在構念向度的假定。本研究針對Mokken量尺分析的步驟與方法、分析軟體取得、分析程式語法、視覺化診斷、題目反應與非參數模型間適配情況進行討論、提供資料詮釋，以及實務應用上的建議。

Mokken scale analysis is a scaling procedure for constructing unidimensional test items. It offers an item selection algorithm to partition a set of items into Mokken scales. It further provides methods to check the assumptions of two nonparametric item response theory (NIRT) models: the monotone homogeneity model (MHM) and the double homogeneity model (DHM). The present study aims to demonstrate applications of NIRT with empirical data. Scalability of data from part of a reading proficiency test for first-year university students (n = 2,270) and from an affective reactions to speaking questionnaire for high school students (n = 168) were examined to illustrate Mokken scale analysis with dichotomous and polytomous items, respectively. It was found that five items on the ten-item reading proficiency test were scalable according to the DHM. On the anxiety questionnaire, only two of twenty items were found to violate the unidimensionality assumption. Our paper also demonstrates use of the open source software program R for unidimensionality and reliability analyses and for graphical diagnostics with the MHM and DHM. We conclude with comments on the process of interpreting results as well as on the practical implications of NIRT modeling for educational and psychological testing.