Estimation of Variances in the Grouped Heteroskedasticity Model
比较了分组异方差模型中两种误差方差估计方法：基于单独组回归残差和基于合并回归残差。蒙特卡洛结果表明，异方差较强时，单独回归残差更有效，合并方法即使迭代也少有显著效率提升。
Error variances for weighted least squares estimation of the grouped heteroskedasticity model can be estimated using residuals from individual group regressions or one pooled regression. The latter is more troublesome, but is usually considered more efficient because the common coefficient vector is imposed on the first stage. However, the variances are biased. Monte Carlo results show this to be an important consideration when heteroskedasticity is strong. Then residuals from separate regressions lead to more efficient weighted least squares estimates. Furthermore, efficiency gains from the pooled method, even when iterated, seldom, if ever, appear to be of large consequence. Copyright 1989 by MIT Press.