中文色塊測驗認知成分分析：LLTM與SEM取向

Abstract

本研究分別以Fischer的線性logistic測驗模式和逐步回歸法對中文色塊測驗進行認知成分分析，並使用結構方程模式檢驗依認知成分加權矩陣建立的試題難度順序結構模式，藉以檢驗中文色塊測驗的建構效度。本研究樣本是依臺灣北、中、南三區六足歲兒童人口比率選取的500位幼兒，主要研究結果有：一、線性logistic測驗模式的認知成分分析結果顯示大小、顏色、方位和順序等四個認知成分，能精簡且有效解釋中文色塊測驗試題之難度變異；二、逐步回歸法得到的認知成分參數估計值與線性logistic測驗模式分析結果相當一致；三、結構方程模式分析結果顯示，實證資料與試題難度順序結構模式適配度良好，受試者在較基礎試題之表現，可預測其在較複雜試題（包括的認知成分多於預測變項）的表現。

This study was aimed at validating the cognitive structures embedded in the Mandarin Token Test (MTT) with Fischer’s linear logistic test model (LLTM) and stepwise regression methodology. Besides, the validation procedure was also based on structural equation modeling of cognitive subordination relationships between test items. Five hundred six-year old children were sampled according to regional distribution of Taiwan’s population. The results showed, firstly, that four major cognitive structures (i.e., size, color, position, and sequence) could adequately explain major variations of the MTT’s item difficulty. Secondly, results from the stepwise regression model were consistent with the LLTM results. Lastly, there was a good fit between the empirical data and the structural equation modeling of cognitive subordination relationships between test items; that is, participants’ performance on lower-level items could successfully predict their performance on higher-level items.