请教一个医学统计问题，谢谢。

stella2011

回复第18楼的 Rtist：

ok, think about a mega clinical trial which conducted in 4 centers, each for 1000 observations.

If these 4000 are correlated with each other with rho=0.1. Then the effective sample size becomes 40 instead of 4000.Using that famular, approximately 40.

Of course the researchers from pharmaceutical industry will happy to claim large sample size 4000 to get a bigger power, or bigger chance to reject null hypothesis. But these 4000 sample size provide no more information than 40 independent observations actually.

stella2011

I'm not intend to annoy you, for such a long discussion.

It do help me a lot, to understand and think more.[s:13]

rtist

回复第20楼的 stella2011：The subject itself is the blocking factor.

rtist

回复第22楼的 stella2011：I'm not annoyed at all, just a little bit sad... I was my mistake to recommend CMH. Shame on me...

回复第21楼的 stella2011：I cannot reproduce a result of 40 using your formula.

Let me try again to persuade you: When we say "information" or effective sample size, we should be more specific to say the "information about xxx". xxx is the parameter that we are interested in. For the same data set, there could be different amount of information on different parameters. However, there are usually some rough lower bound on effective sample size. This can be usually done through counting how many independent blocks in the data. In most cases, the joint distribution for each independent block depends on parameters that are shared by all blocks. Therefore, each block provides at least one piece of information about the underlying common parameter. So the effective sample size for this common underlying parameter is at least the number of independent blocks. This is only a lower bound, because for some parameters, the dependent observations within a block may provide more than one piece of information.

For tedzzx's data, each subject is independent of each other. So the effective sample size should be no less than 33. That's why 3.88 is not a correct answer.

stella2011

回复第24楼的 Rtist：

you are right. [s:13]I should not calculate the effective sample size in that way.

That formular is suitable for cluster data. eg. But in our case ,we assume already independent observation. I'm totally wrong.

rtist

回复第25楼的 stella2011：It is indeed clustered data and the formula could be useful. I think you misunderstood the meaning of <bblatex>n_i</bblatex> and <bblatex>N</bblatex>. As I have said a couple of times, I think you should better plug <bblatex>n_i=2</bblatex> and <bblatex>N=33</bblatex> into the formula.

回复第15楼的 stella2011：On which page have you found this formula? I cannot find it in the book you mentioned. Are you talking about the book by Geert Molenberghs and Geert Verbeke?

stella2011

回复第26楼的 Rtist：

Hi Rtist, I used the formular from Molenbergs' notes which base on that book.

I didn't read the whole book cause we won't go detail this semester. But I can send you that part of note if you are interested or you need a proof:-)

« 上一页