生物醫學研究領域中，當新的藥物或疾病檢測方法提出時，欲了解新的方法是否有相同的治療效果或是不比標準方法差，因此進行等效性或非劣性檢定之評估。為了減少受試者間異質性影響了評估結果，所以可使用成對設計。但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進行成對設計下之等效性與非劣性檢定。此強韌檢定法是將兩個獨立的伯努利概似函數強韌化，得到強韌分數檢定統計量。本文利用模擬和實例分析來展示強韌分數檢定統計量做等效性與非劣性檢定的表現，且將Lu and Bean (1995) 提出的華德 (Wald) 檢定統計量，以及Nam (1997) 基於受限制之最大概似估計建立的檢定統計量來與我們的新方法比較。 ;In medical research, when a new drug or disease detection method is proposed, the equivalence or non-inferiority test is conducted to evaluate whether the new method has the same therapeutic effect, or whether the new method is no worse than the standard method. In this thesis, we propose testing the equivalence and non-inferiority using the robust likelihood function method for paired data. We derive the robust test statistic from the likelihood function constructed by adjusting two independent Bernoulli likelihoods. Via simulations and real data analysis, we demonstrate the performance of our robust procedures for testing equivalence and non-inferiority. We also compare our method to the Wald test statistic proposed by Lu and Bean (1995) and the restricted maximum likelihood estimate test statistic proposed by Nam (1997).
