內外注意係數異常作答診斷空間的建立與探證

Establishing and Examining the Diagnostic Space of Two New Developed Person-Fit Indices: The W(superscript *) and the B(superscript *) Indices

本研究旨在探討內外注意係數的診斷空間關係，利用群集分析技術，以內注意係數(W(上標 *))與外注意係數(B(上標 *))的指標值作為區別受試群診斷空間的空間軸向，並以指標各自均差比(M/SD)作為異常作答歸群的依據，分為內顯、內偏、均等、外偏，與外顯等異常作答類型，最後則配合W(上標 *)與B(上標 *)第90、95與99的百分位數平均決斷值，作為國小數常識測驗實徵資料在指標診斷空間的參照依據。研究結論如下：(1)高能力群偏向於能力內該對卻錯的異常作答，易產生較多內偏型與內顯型的異常作答分配；而低能力群偏向該錯卻對的異常作答，易產生外偏型與外顯型的異常作答分配；至於中能力群則分布在內偏型、均等型與外偏型之間；(2)在與真實資料（高難度試題集）比對的結果顯示，高能力群資料組比較適合二個與四個群集數，低能力群的資料組適合三個群集數的分類，而中能力群資料組則皆適合這三種群集數；(3)結合指標決斷值互相配合解釋，可以進一步有效檢視不同分群類型的異常作答情形的嚴重程度。文末並提供異常作答類型的歸類與使用內外注意係數時的實務應用建議。

The main purpose of this study was to establish and examine the diagnostic-space relationships of two new developed indices， the within-ability-concern index (W(superscript *)) and the beyond-ability-surprise index (B(superscript *)). The diagnostic space expanded by the W(superscript *) and B(superscript *) coordinate axes were established after two runs of cluster analyses. Also after the proportions of mean to standard deviation of W(superscript *) and B(superscript *) indices were calculated， respectively， a ratio of the B(superscript *) proportion to the W(superscript *) proportion were calculated again for setting the five aberrant response patterns (manifest W(superscript *)， slight W(superscript *)， equal-aberrance， slight B(superscript *)， manifest B(superscript *)) on the ratio line from left to right. Finally， combining the 90th， 95th， and 99th percentiles of W(superscript *) and B(superscript *) cutoff scores as the referred points of aberrant response patterns were set for interpreting an empirical data set from the number sense test for elementary school students. Findings showed that (1) high ability students leaned to be classified as the aberrant response clusters of ＂slight W(superscript *)＂ and ＂manifest W(superscript *)＂; low ability students tended to be clustered as the patterns of ＂slight B(superscript *)＂ and ＂manifest B(superscript *)＂; as the medium ability students， a distribution of the aberrant response patterns of ＂slight W(superscript *)＂， ＂equal-strength＂， and ＂slight B(superscript *)＂ were found; (2) in the empirical comparisons， high-ability-high-difficulty (Hh) students were suited for being clustered into 2 or 4 groups， 3 groups were adequate low-ability-high-difficulty students (Lh)， but middle-ability-high-difficulty students (Mh) were suited for all these three groups; (3) it is effective to examine individuals' aberrant response clusters by interpreting the percentiles of W(superscript *) and B(superscript *) cutoff scores. Finally， many suggestions about the classification of aberrant response patterns and the use of the two indices were also provided in text.