J.O.Ramsay 於1991年首先應用既有之核平滑化無參數迴歸函數法,估計試題選項特徵曲線,其能力參數與平滑參數為獨立估計,在能力參數估計部分;劉湘川(民89a)指出其「擴張高低鑑別加權函數」之真實分數估計量值譜數不足,且會發生加權總分逆序情況,另提相關加權估計量以代之,而得改進之核平滑化模式,劉湘川陸續提出系列相關加權改進估計法,本文進而提出「雙重高階相關積累進加權函數」之改進估計法。在平滑參數估計部分;Ramsay之固定平滑參數估計值所得之平滑曲線,其兩尾過度波動,本文提出改進之「動態平滑參數估計值」。在能力參數與平滑參數聯合估計部分;由於核平滑化試題選項特徵曲線常非單調平滑曲線,局部極值出現頻繁,以常用之「條件最大概似估計」不易湊效,本文特發展專有之「條件最大概似數值估計法」,藉以聯合估計無參數試題選項分析模式之能力參數與平滑參數,並以本文發展之「雙重高階相關積累進加權函數」所估得之能力參數值,及「動態平滑參數估計值」,為關聯調適估計之起始值,進行「條件最大概似遞迴數值估計」,據此可得更為有效之改進試題選項分析模式。
The purpose of this study was to provide following three more efficient estimating method for ability parameter and smoothing parameter of item option characteristic curve: (1) Estimating ability parameter: this study provided an improved weighting function, which was combined from higher orders of correlations and a product of higher orders of correlations between the real sample and the perfect sample. (2) Estimating smoothing parameter: this study provided a dynamic smoothing estimator to improve the stability for tow tails of item option characteristic curve. (3) Simultaneous estimating ability parameter and smoothing parameter: this study provided a conditional maximum likelihood numerical estimation for ability parameter and smoothing parameter of item option characteristic curve.