本研究探討配對巢狀二元資料下兩診斷測試比例(如：特異度、敏感度)之 比較問題。在此類資料中，同一病人可能包含多個受測單位，且同一受測單位 同時接受兩種診斷測試，因此資料除具有群聚內相關外，亦存在跨測試間之相 關性，使得傳統方法可能錯估變異數並影響推論結果。為處理此問題，本研究 根據 Royall and Tsou (2003) 之強韌概似推論方法，在實作模型假設獨立下建立 特異度比例之概似函數(likelihood function)，並提出強韌概似比檢定統計量。 本研究進一步推導相關參數之估計量、漸近變異數、修正項與三明治估計 量(sandwich estimator)，並透過模擬研究比較本研究方法與傳統的麥倫瑪檢定 (McNemar test)、調整麥倫瑪檢定、華德類檢定(Wald test)及費雪置換檢定 (Fisher’s permutation test)之表現。模擬結果顯示，當相關性較高或實作模型錯置 時，本研究提出之強韌概似比檢定在型一錯誤(type I error)控制與推論表現上仍 呈現穩定結果。另以實例資料分析結果顯示在考慮資料相關性後，兩種診斷測 試之特異度差異未達統計顯著性。上述結果顯示，本研究所提出之方法可作為 配對巢狀二元資料分析之一種可行工具。;This research considers the comparison of the specificities of two diagnostic tests under clustered paired binary data. In such data， multiple testing units may be observed within the same patient， and each unit is evaluated by both tests， resulting in withincluster and between-test correlations. Ignoring these correlations may underestimate variance and affect statistical inference. To address this issue， we apply the robust likelihood approach of Royall and Tsou (2003) and construct a likelihood function for the specificity ratio under a working independence model. A robust likelihood ratio test statistic is then proposed. We further derive parameter estimators， asymptotic variances， adjustment terms， and sandwich estimators. Simulation studies are conducted to compare the proposed method with the conventional McNemar test， adjusted McNemar test， Wald-type test， and Fisher’s permutation test. The results show that when correlation is high or the working model is mis specified， the proposed method still performs stably in controlling type I error and making inference. Further empirical data analysis shows that， after accounting for data correlation， the difference in specificity between the two diagnostic tests is not statistically significant. These findings suggest that the proposed method is a feasible tool for analyzing clustered paired binary data
