Dr. Frank Yang 问答专帖

思源致远

brainchip 发表于 2013-10-29 15:48
官方DOE Guide Ver9.0，第126页原文如下：
This example shows a power analysis for the Catal ...

可能是typo. 我没有Jmp9 的手册。麻烦你检查Jmp10的同一处，看看是否已更改。要是没有，请告诉我第几页。我仔细看看这部分。

思源致远

brainchip 发表于 2013-10-29 15:56
JMP官方的DOE guide 第9版124页Using the Fit Model Platform提到了一个例题，在对应用Plackett-Burman ...

This JMP example used strong Heredity restriction, which is stronger than the weak heredity that you are familiar with.
On the other hand, if  you keep Catalyst*StirRate in model and you do the regression, you will find out its p value is ~0.1.  You can take it off based this p value. This will end up with the same final model as JMP manual.

brainchip

思源致远 发表于 2013-10-29 22:12
可能是typo. 我没有Jmp9 的手册。麻烦你检查Jmp10的同一处，看看是否已更改。要是没有，请告诉我第几页。 ...

杨老师您好，我看了第十版的同一处，其中改动了delta的值后，原文变成了"The resulting power is
0.8926, which means that in similar experiments, you can expect an 80% chance of detecting a significant
effect for Catalyst."
这一处在官方DOE guide Ver 10.的第142页，chapter 5: A power analysis部分；同一处位于官方DOE guide Ver 9.0的第126页，chapter 4: A power analysis.
谢谢您！

思源致远

Strong Heredity: An interaction is only likely to be significant if both parents are significent.
The p value of Catalyst*StirRate is ~0.1, close to 0.05 on the boarder to take off or keep it in the model.
If you want to keep it, you then better add StirRate back in the model although it is not statistically significant.  When I teach in my class, I suggest to keep them in the model for p value close to 0.1.

brainchip

思源致远 发表于 2013-10-30 22:24
Strong Heredity: An interaction is only likely to be significant if both parents are significent.
T ...

谢谢杨老师的解答！

brainchip

杨老师您好，我还有一个问题，当时看您回复在实验设计版块，我就跟了帖，现在转到您的这个版块来，如下：

思源致远 发表于 2013-10-29 22:48
我找到了废弃没有再用的Jmp9 手册，我的手册上是80% 而不是像你说的89%。不知道你的手册怎么会和我的不一样 ...

我用的是ＪＭＰ９软件Help菜单下books点出的ＤＯＥ　Ｇｕｉｄｅ，其中在１２６页提到的获得的resulting power 是0.802，而检测catalyst效应为显著的几率为89%，如果印刷有误，是不是说明这个resulting power的值 应该和chance for detecting a significance effect的值相等？
谢谢杨老师！

思源致远

I noticed that the resulting power 0.802 is for the model when the interaction Catalyst*StirRate  is taken out of the model.  If you do not take out Catalyst*StirRate considwering its small p value, you then need to add its parent Stir Rate back into model for the sake of strong heredity requirement.  In the later case, you will get 0.8926.
You can try this with JMP's data.  If you still have question, please let me know.  

I need to say a little bit more about Heredity principle.  Some statisticians do not agree such Heredity principles.  However, it is still popularly used by a lot of people.  

思源致远

brainchip 发表于 2013-10-31 10:16
杨老师您好，我还有一个问题，当时看您回复在实验设计版块，我就跟了帖，现在转到您的这个版块来，如下：
...

Yes, you are right. "这个resulting power的值 应该和chance for detecting a significance effect的值相等".

brainchip

思源致远 发表于 2013-10-31 23:00
Yes, you are right. "这个resulting power的值 应该和chance for detecting a significance effect的值 ...

谢谢杨老师！

思源致远

brainchip 发表于 2013-11-1 11:04
谢谢杨老师！

不用客气。希望对你的工作有所帮助。