成對/配對設計引入了相關性/聯合機率，因而使分配間的比較更複雜。本文提出一個強韌概似函數方法來推論配對設計下三個伯努利分配的異同。此強韌法是將三個獨立的伯努利概似函數強韌化，得到強韌分數統計量來檢定三個伯努利分配是否相同。我們並進一步的在模型內加入自變量以考慮配對設計的情境。 本文中提出了理論證明與推導，並利用模擬和實例分析來展示強韌化方法的正確性，以及與非強韌化方法做比較所展示的優勢。;Paired/matched designs introduce correlation/joint probabilities to the probability structure that complicates the analysis considerably. In the thesis, we propose a robust likelihood function approach to inference about the difference between three Bernoulli distributions in paired/matched situations. The robust likelihood is constructed by amending the independent working model. One could acquire a robust score test statistic for testing the homogeneity of three binary populations without modeling correlation/joint probabilities. We further incorporate covariates in the matched scenario. We use simulations and real data analysis to demonstrate the merit of out robust likelihood methodology.
