Rasch模型對項目難易度的估計優於傳統的估計方法乃在於將人的特質以及項目的特質分開考慮, 使用這個模型的優點比起傳統的方法是能夠更貼近現實的情況, 在本篇論文, 主要是採用Rasch模型作為分析工具, 並且將Rasch模型擴展到多個選項的情形。 本論文的第一章主要說明本論文的研究動機以及研究方法, 第二章是預備知識, 主要是介紹用來分析類別資料的多項邏輯模型及對數線性模型, 並說明這兩種模型的相關性, 第三章則是介紹Rasch模型, 並且利用條件概似法把Rasch模型的干擾參數也就是人的特質去除, 這樣的好處是能夠使得項目難易度的估計仍然保有常見樣本的特性, 第四章是介紹分析多選題的部分給分模型, 並且將條件概似法應用在部分給分模型, 最後第五章是結論。 一般常用估計參數的方法是使用迭代的方式, 也因此針對不同的模型必須撰寫不同的程式, 本論文主要是探討條件模型的機率形式類似多項邏輯模型的機率形式, 因此便能夠找出和條件模型所對應的對數線性模型, 這樣使用廣義線性模型的程式就可以把項目的難易度估計出來, 好處是不用針對不同的模型另外寫程式, 更或是購買專門的軟體來分析, 只要利用常見的統計軟體就可以分析本論文的資料。 最後, 針對不同的模型但是其機率形式類似於多項邏輯模型的機率都可以採用本論文的方式來作分析, 並且也希望藉著本論文的研究所探測出項目的難易度, 能夠提供教師在命題時能有所參考。
Compared with the traditional methods, the Rasch model considers the latent trait of human ability and the characteristics of item independently. The advantage of using the Rasch model is that its analysis is closer to the real situation. In this research, the Rasch model is extended to the multinomial case. The method of estimated parameters, which is general used, applies interation. Therefore, it is necessary to design a different program for each model. We show that the functional form of the conditional model is similar to a multinimial logit model which has an equivalent loglinear model form. We have found that the functional form of model similar to that of uaual multinomial logit model. Therefore, the item difficulty can be estimated by using the generalized linear modeln program. The advantage of using linear model is that it does not require a new software or program to analyze the data, i.e. general statistial software can be used. Moreover, the method discussed in this research also can be used to analyze different models which the functional forms are similar to that of multinomial logit model. Finally, We hope the item difficulty calculated by the research will help teachers when they design tests.