Discussion on the paper by Professor Wu, "A fresh look at effect aliasing and interactions: some new wine in old bottles"
Discussion on the paper by Professor Wu, "A fresh look at effect aliasing and interactions: some new wine in old bottles"
复制标题
对吴教授论文的讨论，“A fresh view oneffectaliasingandinteractions:somenewwineinold Bottles”
DOI：
10.1007/s10463-017-0641-x
复制
发表时间：
2018
影响因子：
1
通讯作者：
Ryo Yoshida
中科院分区：
文献类型：
--
作者：
Trung Anh Dinh;Shigeru Yamashita and Tsung-Yi Ho;Ryo Yoshida
我对吴教授及其合著者在统计科学史上取得的伟大成就表示敬意。在传统的实验设计中，因子交互作用通常是别名的，如 24−1 设计中别名的双因子交互作用所示。事实证明，吴教授论文中基于方程 3 或方程 4 的 CME 重新参数化可用于消除常规 2k−q 设计中混叠交互效应的混叠。正如吴教授所讨论的，CME 分析不仅在设计实验中具有很大的适用性，而且在观测研究中也具有很大的适用性。让我们关注 A 和 B 之间的双因素相互作用，每个因素都有两个水平，+ 或 -，表示相应因素的存在或不存在。传统上，交互效应是量化A和B对响应变量的产品类型影响，其定义为相同符号（都存在或不存在）和相反符号状态之间的平均效应之差。这仅描述了更广泛背景下交互的一个方面。 CME 为我们带来了对相互作用的另一种看法，它在许多应用中提供了更有意义的科学见解。 Mak 和 Wu（2017）在观察研究中开发了 CME 分析的综合框架，并采用基于效应分组的变量选择程序（Mak 和 Wu 2017）。
I express much respect to the great achievements in history of statistical science that have been made by Professor Wu and his coauthors. In conventional experimental design, the factor interactions are often aliased as exemplified for the aliased two-factor interactions in the 24− 1 design. It has been shown that the CME reparameterization based on Eq 3 or Eq 4 in Wu’s paper could be used to de-alias the aliased interaction effects in regular 2k− q design.As discussed by Professor Wu, the CME analysis has the great applicability not only in designed experiments but also in observation studies. Let us focus on the twofactor interaction between A and B with each having two levels,+ or−, that indicates the presence or absence of the respective factor. Conventionally, the interaction effect is to quantify the product-type influence of A and B on a response variable, which is defined to be the difference of the mean effects between the same signed (both are present or absent) and opposite-signed states. This describes merely one aspect of the interaction in a broader context. The CMEs bring to us another look on the interaction, which provide scientifically more meaningful insights in many applications. Mak and Wu (2017) developed a comprehensive framework of the CME analysis in observation studies with a variable selection procedure based on the effect grouping (Mak and Wu 2017).