摘要 二元資料之分析大部分的情形都是使用邏輯斯迴歸模型。但事實上，適合之模型不見得是邏輯斯迴歸模型。 本文之目的在使用Royall and Tsou(2003)之強韌概似函數法。將二項分配模型做適當之修正後，使得在大樣本之情形下，即使二元資料之分配非二項分配，亦能對連結函數做正確之推論。 Abstract This thesis utilizes the robust likelihood technique proposed by Royall and Tsou(2003) to develop parametric robust inferences about the link function for binary response variables. More specifically, the binomial model is corrected to become robust. With large samples the adjusted binomial likelihood is asymptotically legitimate for the relationship of interest. One needs not know whether the binary data are correlated or not, or the extent of the correlation exists in data. Simulations are used to demonstrate the efficacy of the proposed robust method.
