Preliminary-Test Estimation of the Error Variance in Linear Regression
推导了线性回归模型中几种常见事前检验估计量的偏差和风险的精确有限样本表达式，发现最小均方误差成分估计在典型情况下表现最优。
We derive exact finite-sample expressions for the biases and risks of several common pretest estimators of the scale parameter in the linear regression model. These estimators are associated with least squares, maximum likelihood and minimum mean squared error component estimators. Of these three criteria, the last is found to be superior (in terms of risk under quadratic loss) when pretesting in typical situations.