j*********1
发帖数: 21 | 1 如题。感觉principal component analysis 在我这个field还算比较新的方法。不知道
在座牛人有没有做过他的power analysis. 请问它有比较conventional 的power
analysis方法吗？ Pca算出来后打算放进regression model里面。是不是只能做
regression model的power analysis啊。
新人实在没有包子，望各路牛人不吝赐教。小女子这厢有礼啦. |
b*****n
发帖数: 685 | 2 啥叫PCA的power analysis？你不做test哪来的power？ |
j*********1
发帖数: 21 | 3 谢谢楼上牛人。test就是multiple regression。整个model大概是Y=PCA index+
covariates
我的意思是，创建PCA index里面用到的variables，他们的sample size会影响到整个
model的power吗？
刚接触PCA，理解还不是很透彻，问了菜鸟问题多包涵哈。：）
【在 b*****n 的大作中提到】
: 啥叫PCA的power analysis？你不做test哪来的power？
|
D**u
发帖数: 288 | 4 I think when you say "power of model" you really want to say the "goodness
of fit" of that model.
Power is only of a hypothesis testing, a model is not a hypothesis testing,
and multiple regression is a model, so there is no "power" related to
multiple regression. However, there are a lot of tests of a multiple
regression model's "goodness of fit" available for you, and you can always
say a model is "powerful" when it fits well.
And if above is what you meant, then a quick answer to your question is "Yes
". As you will always find in almost all testings, power is related to
sample size, and goodness of fit tests are no exceptions no matter PCA or
anything else you put in them. |
w*******9
发帖数: 1433 | 5 "There are a lot of tests of a multiple regression model's "goodness of fit"
." I dont think goodness of fit tests for Linear Regression are developed as
well as their counterparts in Logistic Regression. Maybe I am wrong? |
j*********1
发帖数: 21 | 6 Thank you for your explanation. If I understand correctly, in a multiple
regression model, the hypothesis test is done by either T test or Chi-square
test, right? What I meant is when you are doing a power calculation to a
hypothesis test in a a multiple regression model that included one of the
predictors is a PCA index, is there anything special to it or we can just
treat it as a simple continuous variable? Sorry about the bad wording
earlier. I am not a professional statistician. :P
I think what you mentioned in your post might be different from what I was
looking for. I think the model fit test is quite different from power
analysis. But thank you none the less. :)
,
Yes
【在 D**u 的大作中提到】
: I think when you say "power of model" you really want to say the "goodness
: of fit" of that model.
: Power is only of a hypothesis testing, a model is not a hypothesis testing,
: and multiple regression is a model, so there is no "power" related to
: multiple regression. However, there are a lot of tests of a multiple
: regression model's "goodness of fit" available for you, and you can always
: say a model is "powerful" when it fits well.
: And if above is what you meant, then a quick answer to your question is "Yes
: ". As you will always find in almost all testings, power is related to
: sample size, and goodness of fit tests are no exceptions no matter PCA or
|
j*********1
发帖数: 21 | 7 I think there are many goodness of fit tests for Linear regression model.
Like Chi-square, residual plots, etc., you can even use lovely R square for
part of your diagnosis in a linear regression model. For me, that's plenty.
:)
fit"
as
【在 w*******9 的大作中提到】
: "There are a lot of tests of a multiple regression model's "goodness of fit"
: ." I dont think goodness of fit tests for Linear Regression are developed as
: well as their counterparts in Logistic Regression. Maybe I am wrong?
|
D**u
发帖数: 288 | 8 OP, u must be trolling...
for
.
【在 j*********1 的大作中提到】
: I think there are many goodness of fit tests for Linear regression model.
: Like Chi-square, residual plots, etc., you can even use lovely R square for
: part of your diagnosis in a linear regression model. For me, that's plenty.
: :)
:
: fit"
: as
|
j*********1
发帖数: 21 | 9 I don't get it... what did you mean?
【在 D**u 的大作中提到】
: OP, u must be trolling...
:
: for
: .
|
h***x
发帖数: 586 | 10 Troll 作动词是钓鱼的意思
Troll 的网络意义: 公共论坛用侮辱性言语挑起骂战并从中获取某种不知名快感的人
台湾人称“小白”，国内称“脑残”
表怪我，楼上Dinu 说的，俺也不知道是啥意思，只是好奇，google了帮您翻译一下...
【在 j*********1 的大作中提到】
: I don't get it... what did you mean?
|
|
|
D**u
发帖数: 288 | 11 这个解释不是我本意啊...楼上害我，我就是想开个玩笑啊...troll很多意思啊，也可
以说成是挖坑啊，故意说反话，装小白啊之类的，是夸楼主是有智慧的萌主啊
楼主你要是有啥问题可以直接就问我好了，我会的就帮你回答了
..
【在 h***x 的大作中提到】
: Troll 作动词是钓鱼的意思
: Troll 的网络意义: 公共论坛用侮辱性言语挑起骂战并从中获取某种不知名快感的人
: 台湾人称“小白”，国内称“脑残”
: 表怪我，楼上Dinu 说的，俺也不知道是啥意思，只是好奇，google了帮您翻译一下...
|
D**u
发帖数: 288 | 12 I am not a expert in the knowledge of testing, just my .02, no guarantee to
be right.
If you want to test for goodness of fit of a linear
regression model. The first one coming to my mind is F-test, and there are
tests like Kolmogorov–Smirnov test could be used also.
And things like residual plots, R square, AIC, DIC... are statistical
measures of the quality of a model, they are not testings, so no power
related to them.
for
.
【在 j*********1 的大作中提到】
: I think there are many goodness of fit tests for Linear regression model.
: Like Chi-square, residual plots, etc., you can even use lovely R square for
: part of your diagnosis in a linear regression model. For me, that's plenty.
: :)
:
: fit"
: as
|
w*******9
发帖数: 1433 | 13 Generally GOF test requires replicated observations, you need to group
somehow. Unlike the logistic regression, for linear regression, there's no
universally accepted approach for grouping. If U google, you can only find
very few GOF tests for the general linear regression and non of them are
found in the textbook.
to
【在 D**u 的大作中提到】
: I am not a expert in the knowledge of testing, just my .02, no guarantee to
: be right.
: If you want to test for goodness of fit of a linear
: regression model. The first one coming to my mind is F-test, and there are
: tests like Kolmogorov–Smirnov test could be used also.
: And things like residual plots, R square, AIC, DIC... are statistical
: measures of the quality of a model, they are not testings, so no power
: related to them.
:
: for
|
h***x
发帖数: 586 | 14 I was joking as well. :-)
【在 D**u 的大作中提到】
: 这个解释不是我本意啊...楼上害我，我就是想开个玩笑啊...troll很多意思啊，也可
: 以说成是挖坑啊，故意说反话，装小白啊之类的，是夸楼主是有智慧的萌主啊
: 楼主你要是有啥问题可以直接就问我好了，我会的就帮你回答了
:
: ..
|
j*********1
发帖数: 21 | 15 哈哈，我自认小白～background就不是统计，只是半路出家修了一些课，所以
statistic theory那块都一片空白～～要是挖坑的话，我觉得此坑故事性略欠了点～～
还是回到我的问题，如果是普通regression model的power analysis还好，做个Monte
Carlo simulation就可以了。我就是concern如果有predictor是PCA做出来的index,有
没有什么东西需要特殊处理的。上网搜也搜不到什么～～呵呵
【在 D**u 的大作中提到】
: 这个解释不是我本意啊...楼上害我，我就是想开个玩笑啊...troll很多意思啊，也可
: 以说成是挖坑啊，故意说反话，装小白啊之类的，是夸楼主是有智慧的萌主啊
: 楼主你要是有啥问题可以直接就问我好了，我会的就帮你回答了
:
: ..
|
j*********1
发帖数: 21 | 16 Ahh, I see. Thank you.
to
【在 D**u 的大作中提到】
: I am not a expert in the knowledge of testing, just my .02, no guarantee to
: be right.
: If you want to test for goodness of fit of a linear
: regression model. The first one coming to my mind is F-test, and there are
: tests like Kolmogorov–Smirnov test could be used also.
: And things like residual plots, R square, AIC, DIC... are statistical
: measures of the quality of a model, they are not testings, so no power
: related to them.
:
: for
|
a*********9
发帖数: 14 | |
f****s
发帖数: 3078 | 18 第一次听说power analysis...
【在 j*********1 的大作中提到】
: 如题。感觉principal component analysis 在我这个field还算比较新的方法。不知道
: 在座牛人有没有做过他的power analysis. 请问它有比较conventional 的power
: analysis方法吗？ Pca算出来后打算放进regression model里面。是不是只能做
: regression model的power analysis啊。
: 新人实在没有包子，望各路牛人不吝赐教。小女子这厢有礼啦.
|
T*****u
发帖数: 7103 | 19 are you trying to figure out how PCA affects variation and bias in linear
regression? |
y**3
发帖数: 267 | 20 power analysis is to calculate sample size, is part of work of Experimental
design, as my understanding |
