科研之友
微信
新浪微博
Facebook
分享链接
摘要
Clinical trials and many other scientific studies perform statistical tests for multiple response variables for finding whether an experimental treatment of interest is better than a specified control for some of these variables. Multiplicity adjustments for such tests of multiple response variables, having varying degrees of correlations, have been a challenging task for achieving a meaningful control of the Type I error rate. A way out of this difficulty has been simply to ignore correlations among response variables and use methods such as the Bonferroni procedure or the. Sidak (1967) adjustment formula under independence. This, however, can make the multiplicity adjustments conservative when the correlations are in the moderate to high range and reduce the power of the tests for finding beneficial treatment effects. On the other hand,. Sidak's adjustment formula under independence is quite attractive; it's easy to use for multiplicity adjustments and for setting simultaneous confidence intervals. Therefore, there has been interest in ad-hoc modifications of this formula that incorporate correlation information for making it less conservative and for gaining power, but at the same time retaining its simplicity. There have been several attempts in this regard but without much success. We review prior attempts at such modifications and propose a new one which is easy to use and controls the family-wise error rate nicely. An attractive feature of the proposed method is that its tests easily convert to simultaneous confidence intervals-most popular methods, such as step-up and step-down type methods lack this property.
- 出版日期2010-2
