本文利用強韌概似函數方法，在巢狀群集離散資料結構下，提出兩種檢驗處理效果間差異的強韌概似推論法。我們將根據兩個獨立多項式分配實作模型得到分數統計量強韌化，根據此強韌化的強韌分數統計量可在不需要針對巢狀群集間的相關性建構模型情況下得到正確的，有關兩種檢驗處理效果間差異的統計推論。我們也提出參數的概似比檢定統計量，及據此建構的信賴區間。我們用模擬來展示強韌概似方法的優點。;We establish robust likelihood inference for the effects difference between two treatments in the scenario of correlated categorical data with nested structure. More specifically, we robustify the naïvecore test statistic based on two independent multinomial working assumptions. The robust score test statistic is able to deliver legitimate inference without the knowledge of the correlation structure. In addition, robust likelihood ratio statistic and the test-based confidence interval are also provided for the parameter of interest. We use simulations to demonstrate the efficacy of our parametric robust method.
