Rasch模式建置「國小四、五年級閱讀 理解測驗」簡版

Development of a Shortened Reading Comprehension Test Using the Rasch Model

國內目前根據閱讀認知歷程，專門針對國小一般學童編製之文本 閱讀理解測驗仍有限，在少數測驗中，測驗長度又限制了實務上的應 用。本研究主要目的在以Rasch部份給分模式縮減依據NAEP閱讀評量 架構所編製之「國小四、五年級閱讀理解測驗」，並透過比較縮減後簡 版與原版在各項心理計量特性上的差異，來評估簡版的適切性。研究資 料取自黃琬玲（2010 )及張麗麗、陳品華、羅素貞、陸怡琮（2011 )， 共計1914位高、屏地區41所國小之四、五年級學童。本研究以Rasch 模式建置及評估簡版適切性的原則包括：簡版試題在涵蓋範圍（兩種閱 讀文本及三個認知層次）、廣度及與受試者能力對應上需要與原版相 符；試題長度適合於一節課內施測完畢；簡版在試題難度、向度性（等 級量表適切性、試題契合度、性別/年級/學校規模DIF )、能力估計恆 定性、信度等各方面需與原版相當。結果顯示在縮減原版測驗達41% 幅度及施測時間為一半的情形下，簡版仍維持與原版相同或甚至更理想 的計量特性。本文亦針對Rasch模式在縮減測驗上的應用、後續研究建 議及測驗應用提出討論。

Although recent research in reading comprehension has focused on the processes of reading， only a few such reading comprehension tests are available for primary school children in Taiwan. Of the few available tests， the length of the test further limits the extent to which they actually applied in school settings. Therefore， the present study aims to shorten and validate a reading comprehension test developed， based on the NAEP reading assessment framework for 4th and 5th grade students using the Rasch partial credit model. Two data sets， collected separately by Hwang (2010) and Chang， Chen， Lo， and Lu (2011)， consisting of a total sample of 1，191 4th and 5th graders from Kaohsiung and Pingtung areas in Taiwan， were used to develop and evaluate the psychometric properties of the shortened reading comprehension test. The adequacy of the shortened test was assessed with respect to the content coverage of the test (including two types of reading passages and three levels of cognitive processes)， item spread and item-person alignment， the dimensionality of the test (including rating scale performance， item fit to the model， item calibration invariance [DIF] across gender， grade and school size)， the invariance of ability estimates across forms， and person separation reliability. The results indicate that the shortened version， with a reduction of 41% of the items and half of the administration time， achieves similar or even higher quality of the psychometric properties as compared with that of the original version. Discussions on test reduction using the Rasch model， suggestions for further studies， and applications of the reading comprehension test are also provided.