在判別多組連續型資料的變異係數是否相等時，常假設資料服從常態分配，然後進行分析。但是，當資料不服從常態分配時，那麼根據此假設所得到的統計推論便會是不對的。 本文將Royall and Tsou(2003)提出的強韌概似函數的方法，應用在多母體變異係數間之比率的推論上。利用經適當修正過的常態概似函數，提供一個只要母體真正分配的四階動差存在，變異係數正確的統計推論方法。 This paper uses the robust likelihood technique proposed by Royall and Tsou(2003) to develop a parametric robust score test for testing the equality of coefficients of variation. More specifically, it is demonstrated that the normal likelihood function could be adjusted to provide asymptotically valid inferences for practically all continuous random variables, as long as the underlying distributions have finite fourth moments. Three-sample problem is investigated in details, and the adjustment that achieves the robustness property is presented. Simulation studies are executed to demonstrate the finite sample performance of the novel robust procedure. Real data analyses are provided to compare several procedures for testing the equality of coefficients of variation.
