二元計分的BW作答機率模式精確度研究：蒙地卡羅法

The Accuracy of Dichotomous BW Probabilistic Models: A Monte Carlo Study

本研究旨在探討以異常作答偵測指標BW內外注意係數所建構的兩種二元計分BW作答機率模式：BW內嵌模式（BWEM）與BW本位Rasch模式（BWRM）的實用性，期待經由檢視模式作答預測能力的精確度，為模式應用增添可靠的支持。本研究以命中率代表模式的作答預測能力，並作為評估模式的準則。透過比較BWEM與BWRM在不同受試者人數與測驗長度下的命中率，探討兩種作答機率模式於實務應用時需留意之處。透過三因子混合設計變異數分析，研究結果顯示：（1）受試者人數、測驗長度及作答機率模式三者間的確存在交互作用，但是效果量很低；（2）整體、勝券在握與蜀道難登三區，命中率在不同受試者人數、測驗長度及作答機率模式間的確存在顯著差異；（3）BWEM與BWRM的命中率孰高孰低，視所在的作答反應區域而定。

The main purpose of this study was to examine the accuracy of two dichotomous BW probabilistic models constructed by BW aberrance indices: the BW embedded model (BWEM) and the BW based Rasch model (BWRM). This study used the hit rate as an index to evaluate the accuracy of the models. Through examining the differences between hit rates of BWEM and BWRM under different circumstances (number of examinees, test length, and probabilistic model), this study explored the characteristics of these two probabilistic models. The results from the three-way mixed design ANOVA showed: 1. There exists a three-way interaction effect among the number of examinees, test length, and probabilistic model, but the strength of association was not strong. 2. There existed significant differences of hit rates among specific response areas in the conditions of number of examinees, test length, and probabilistic model, respectively. 3. The hit rate of BW embedded model could be higher or lower than that of BW based Rasch model, depending on which response area is concerned.