Robust Multiple Comparisons
研究了单因素布局中位置参数的学生化M估计的联合Edgeworth展开，比较了Bonferroni、最大模和Scheffé三种多重比较方法的稳健性，发现前两种在比较次数多时不稳健，Scheffé方法更稳定。
Abstract The first-order terms of the joint Edgeworth expansion for the p Studentized M estimates of location in a oneway layout are presented and the probabilities that generate the Bonferroni, maximum modulus, and Scheffé methods of multiple inference are approximated. The validity-robustness of these methods, based on both classical estimators and a variety of M estimators, will be compared for increasing numbers of simultaneous comparisons. The Bonferroni and maximum modulus methods are seen to be nonrobust when many comparisons are made. Scheffe's method is more stable. These results are fairly independent of the choice of estimator. Monte Carlo results indicate these expansion approximated probabilities accurately reflect qualitative behavior and give reasonable numerical values for moderately trimmed M estimators, but are insufficiently accurate to generate critical points for heavily trimmed estimators.