对稳健回归尺度参数估计的一种改进

An Alternative Estimation for the Scale Parameter in Robust Linear Regression Models

常对线性回归模型的稳健 M估计中 ,尺度参数使用绝对离差中位数 MAD.将 Rousseeuw等人对单变量尺度参数的一种稳健估计 Sn引入到回归问题中 ,讨论了此估计的一些优良性质 ,并通过一个小规模的模拟研究 ,说明使用 Sn比使用 MAD做尺度参数将会较大地提高回归估计的估计效率 .

In robust M estimations for linear regression models, median absolute deviation often used as the scale parameter estimator. In this paper, we introduce another robust scale estimation originally proposed by Rousseeuw under univariate situation into linear regression. Some property about this scale estimation is mentioned and a limited Monte Carlo study conducted. It shows that this robust scale estimation can enhance the estimate efficiency of regression coefficients for Huber′s M estimators.