广义线性模型标准误折刀法稳健估计的模拟研究

Simulation study of the Jackknife method for the robust standard error estimation in generalized linear models

探讨广义线性模型中标准误折刀法估计的不同形式,通过蒙特卡洛模拟考察了两种基于折刀法标准误估计之表现.模拟结果表明,折刀法给出的标准误在模型设定正确时与基于Fisher信息阵的标准误行为类似,而在独立性假定被破坏时,折刀法明显优于后者.最后,将折刀法用于分析一组癫痫病数据,得到了令人信服的结论.

Within the framework of the Jackknife method, two robust estimations of the standard error of the quasi - maximum likelihood estimator in the generalized linear model are discussed. Monte - Carlo simulations are conducted to assess the performances of this method. The simulation results show that the performances of the Jackknife method are comparable to the standard error derived from Fisher information matrix （i. e. FO method）, when the independent assumption holds. If the assumption fails, the Jackknife method apparently outperforms the FO method. The Jackknife method is adopted to analyze a set of epileptic cases, and convincing results are obtained.