送交者: xj 于 2005-8-21, 01:07:22:
回答: XJ, do you know how to do power analysis 由 cornbug 于 2005-8-21, 00:53:20:
I am not sure what your question is. If you ask for sample size before experiments, that is the sample size calculation. It is based on your estimated effect difference, alpha (0.05-0.01), power (0.8-0.95), your estimated standard error, and the proposed test itself. what kind of tests you will use in your problem: t test? chisquare test? linear regression based with lots of adjustment? logistic regression or suvival analysis? They are different.
Given a sample size, you can do a power analysis. What are the P values for type 1 and 2 errors for each estimated difference?
we usually give out two tables for these two things.
There is a software called NCSS/Power which you can download a trial version from its website. However, most people do it by hand. The issue is what the problem you have.
Power analysis is very tricky and you'd better ask some competent people to do that--most students probably don't know the details.
I may know some, but others I have to look into my books.
Are you writing a grant, or just a post hoc estimation in a paper?