在模型配適過程中，機差(deviance)扮演著判斷模型是否合適的重要統計量，此統計量反應了資料與假設的模型間的差異。而根據定義，機差是依據使用的分配不同而不同。本文將試著建立在資料來源分配不知道的情況下具強韌性質的機差以用來確認迴歸模型的配適是否合適。;Deviance plays a crucial role in model checking. It is the scaled difference between the model entertained and the so-called saturated model. Obviously, by definition deviance is model-dependent. This research explores the possibility for constructing less model-dependent or robust deviances that can be used for checking the appropriateness of a regression specification without knowing the true underlying distribution.
