基于Fisher信息阵的局部影响分析

建立了参数空间中以Ｆｉｓｈｅｒ信息阵为Ｒｉｅｍａｎｎ度量的距离，应用到回归模型中，得到了观察数据的影响度量，讨论了方差扰动模型，因变量扰动模型以及自变量扰动模型的局部影响，求出了因变量扰动模型的最大曲率方向，以一个实例与从似然距离出发的局部影响作了比较，结果表明基于Ｆｉｓｈｅｒ信息阵的影响度量是合适的。

The Loeal influence Asalysis Based onthe Fisher information Matrix

A new distance based on the Fisher information matrix is proposed in the parametricspace, and it can be used to measure the influence of individual obervation and perturbation in regres-sion model. Furthermore, the case-weights perturbation model, response variable petwrbation modeland explanatory variable perturbation model are discussed and the maximum curvature direction corre-sponding to perturbing response variable is given. The diagnostics are illustrated by the comparesion ofan example with the local influence analysis based on the likelihood displacement.