生物醫學研究領域中，當新的藥物出現時，欲了解新的藥物與舊的藥物是否具有相同的治療效果或不同程度的不良影響，常會使用成對設計。若在多個疾病與成對設計下，則使用多重成對設計，但多重成對設計引入的相關性使模型配適變得困難。 本文提出利用強韌概似函數方法來來分析多重成對的二元資料。此強韌檢定法是將多個成對且獨立的伯努利概似函數強韌化，並得到強韌分數檢定統計量。本文利用模擬與實例分析比較強韌分數檢定、Lui and Chang (2013) 提出的修正華德 (Modified Wald) 檢定統計量及 Klingenberg and Agresti (2006) 提出的Multivariate extensions of McNemar’s test。;In medical research, when a new drug is proposed, in order to understand whether the new drug and placebo have the same treatment/adverse effects or different extent of adverse event, we use the matched-pair design. If the patient has multiple outcomes, we have the matched-pair design with multiple endpoints. The within-pair and between endpoints correlations make likelihood inference extremely difficult. In this thesis, we propose a robust likelihood approach to analyzing paired multiple binary endpoints data. We use simulation and real data analysis to demonstrate the merit of our new parametric robust technique. We also make comparison with to the Modified Wald’s test procedure proposed by Lui and Chang (2013) and the multivariate extensions of McNemar’s test statistic proposed by Klingenberg and Agresti (2006).
