隨機邊界法應用在縱橫資料上往往將廠商的無效率項以固定效果或隨機效果項代替，然而此方法會將代表廠商差異性的異質效果包含至無效率項。為了解決此問題學界已有論文推導出可分離廠商無效率項與廠商異質項的方法，然而該方法存在著Incidental parameters problem，後來的文獻藉由一階差分法以及組內轉換法先排除廠商異質效果，再由一般的最大概似估計法估計相關的待估計參數，最後再由估計得出參數估計值由殘差的方式得出廠商的異質效果項，這樣的好處是可以避免Incidental parameters problem。然而上述提及的方法均假設組合誤差項內元素彼此間獨立，學界尚無明顯的證據支持該假設，故本論文將提出一套組合誤差項內元素彼此相關的模型，包括了未分離廠商異質效果與無效率項的組合誤差項內元素不獨立模型以及分離廠商異質項與無效率項的組合誤差項內元素不獨立的固定效果模型。 The method of stochastic frontier analysis with panel data used in the field of productivity and efficiency considers firm’s heterogeneity as individual inefficiency term in the past. The problem of this approach is that the inclusion of firm’s heterogeneity in its inefficiency term has not accessible ability to gain precisely the information of pure inefficiency from fixed effect or random effect. In order to get rid of this troublesome weakness, a model with capability to separate firm’s heterogeneity and inefficiency term is developed .However, This model also arises the incidental parameters problem. In order to immune from this problem, the usages of first difference and within group transformation are worked to eliminate heterogeneity and estimate other parameters that is remained in the model before practicing the method of maximum likelihood estimation. Another issue is that models mentioned above are all based on the assumption of independent composed error. Because the lack of evidence for assuming independent composed error for the methodology of stochastic frontier analysis in the field of productivity and efficiency, this prompts us to build two stochastic frontier models with the hypothesis for dependent composed error. The first is going to derive a model without separating firm’s heterogeneity from its inefficiency term. The second is a fixed effect model with a heterogeneity term. Furthermore, the assumption of dependent composed error is imposed on both models.