Lyles et al. (2007)提出一個針對廣義線性模型，利用資料擴充來估計統計檢定力的方法。由於此方法需要假設反應變數的分配，再利用費雪訊息矩陣估計變異數矩陣，並利用Wald 檢定統計量之漸進分佈估計統計檢定力。但當模型假設錯誤時，所得到的費雪訊息矩陣基本上是不正確的。本文之目的在於利用拔靴法來估計當模型假設錯誤時正確的變異數矩陣，再利用Wald 檢定統計量之漸進分佈估計出正確的統計檢定力。 Lyles et al. (2007) proposed an expanded data set method for calculating testing power in the setting of generalized linear models. This approach requires the Fisher information matrix in order to evaluate the Wald test statistic. We recommend using the Bootstrap methodology to calculate a robust version of the Fisher information matrix which remains legitimate under model misspecification. Hence, one can estimate the power of the test statistics without making distributional assumptions.
