具有一般协方差结构线性模型的局部影响评价

该文讨论了具有一般协方差结构线性模型的局部影响分析问题. 通过对广义Cook统计量中M/c的适当选取, 文章给出了一种对扰动的参数变换具有不变性质的局部影响度量. 在具协方差扰动模式下, 该文给出了回归系数和方差系数估计、最佳线性预测的局部影响诊断统计量.该结果与数据删除法进行了比较, 并通过实例进行了分析和说明.

Local Influence Assessment in the General Linear Model with Correlated Errors

Assessment local influence analysis in linear model with correlated errors is studied in this paper. The method is based on generalized Cook statistic (Shi, 1997) with a suitable choice of M/c such that the local influence measure is invariant under parameterization. A perturbation scheme is introduced through covariance matrix. The local influence measures for the estimators of regression coefficients, variance coefficient and the best unbiased linear prediction are derived. The results are compared with those of case-deletion, and a concrete example is used to illustration.