在處理二維個數資料時，為了分析時的便利，多數情況下都會假設資料服從二維卜瓦松分配或二維負二項分配。但是一旦資料不是來自二維卜瓦松分配或二維負二項分配時，那麼根據二維卜瓦松分配或二維負二項分配模型所做的統計推論便是錯誤的。 本文將Royall and Tsou(2003)所提出的槪似函數修正法應用於二維個數資料之母體平均數之比率的推論上，說明所提出的二維卜瓦松實作模型和二維負二項實作模型是可以經過適當的修正而被強韌化。在大樣本時，且部分正規條件下，不論資料真正分配為何，據此強韌槪似函數可對母體平均數之比率參數做正確統計推論。 This thesis utilizes the robust likelihood technique proposed by Royall and Tsou (2003) to develop parametric robust inferences about the comparison of two dependent populations of counts. More specifically, bivariate Poisson and bivariate negative binomial models are corrected to become robust. With large samples the two adjusted likelihood functions are asymptotically legitimate for the parameter of interest, without the knowledge of the true underlying distributions. Simulations are used to demonstrate the efficacy of the proposed robust method.
