國家教育研究院

　　本研究是一年期計畫。主要是探討結合知識結構及Q矩陣之認知診斷測驗編製和成效分析；認知診斷測驗常透過Q矩陣表徵試題和認知屬性之關係，目前對於認知診斷測驗的研究仍是以無結構性的Q矩陣設計為主，本計畫以知識結構為基礎編製包含選擇題和建構反應題型之認知診斷測驗，將知識結構之階層性納入Q矩陣設計中，探討不同的Q矩陣設計對於認知診斷模式之影響。模擬研究結果顯示：無結構性Q矩陣時，隨著Q矩陣包含單一認知屬性試題的題數越多，不同認知診斷模式的猜測度與粗心度估計誤差均越小且受試者認知屬性正確分類率的估計精準度越高。對於結構性Q矩陣，需根據不同結構，設計不同的形式試題，方能得到較佳的估計效果。實證資料研究顯示：結合專家知識結構之Q矩陣設計具有較佳的估計效果。

　　This study is proposed for one-year period of research implementation. The project focuses on developing cognitive diagnostic assessments combining knowledge structures in the Q matrix designs and evaluating its performance. Implementation of the cognitive diagnosis models requires construction of a Q matrix to describe how the test items are related to the attributes. Many applications of cognitive diagnosis are based on the assumption that attributes are independent； however, it is more reasonable to assume that attributes are dependent and follow some type of structure in some application. In this study, based on knowledge structure a cognitive diagnostic assessment with both multiple choice and constructed response items has be developed. The effects of different designs (independent and structured) on cognitive diagnosis models were explored. In the simulation study, under the independent Q matrix designs, with increasing of one-attribute items, the estimation errors on item and attribute parameters decrease in diagnostic models； under the structured Q matrix designs, the performance of attribute and item estimation depended on different structured Q matrix design With two types of the Q matrixes, the performance of the expert knowledge structure was better than that of the student knowledge structure. In the empirical study, with two types of the Q matrixes, the performance of the expert knowledge structure was better than that of the student knowledge structure.