05-09-2017 02:36 PM
If I know the population size (such as 10,0000 records), and I set the confident level to be 90%, upper error limits to be 7%, expected error rate to be 2%. Can I use proc power to calcuate the sample size. It seems to me this is an attribute sample size determination, not parametric one.
Thanks for your help.
05-09-2017 02:54 PM
Possibly but you'll need to provide the type of statitical test you are going to do and what you are actually testing. Your post is kind of short of details.
05-09-2017 04:18 PM - edited 05-09-2017 04:20 PM
Thanks! What I am trying to do is to decide sample size by specifying population size (such as 10,000 claim records), confident level, Upper Error Limits, and Expected Error Rate. If I do random sampling twice based on the calculated sample size (sample pool A and sample pool B), what I find the pattern in sample pool A can be also found in sample pool B.
05-11-2017 09:40 AM
What ballardw is saying is that determining the sample size that is required to get a significant result with a specified probability DEPENDS on the analysis that you are conducting. Are you planning to do an ANOVA? A two-sample t-test for means or proportions? Tell us what statistical test you plan to run.
You might like to read this overview of power and sample size computations, which gives a lot of useful background information.
05-11-2017 02:01 PM
Thanks for your reply. My sample size determination is used for audit, not for hypothesis testing, for example, if there are 100,000 claim records, and I would like to find out what is the billing pattern. If 1,000 is calculated as the sample size, then I hope what I find the billing pattern from this 1,000 records is also true to another 1,000 sample drawn from the same 100,000 population. So there is no specific variable I want to test. I hope I put it clear this time.
05-11-2017 02:23 PM
You keep saying the same thing, but the example you give doesn't seem to conform to the usual definition of sample size determination.
I don't think it is possible to do what you are asking for. It sounds like you want to find a sample size N such that you can test future samples of size N (non-parametrically?) against an original sample and conclude that the distributions come from the same population.Testing whether two samples come from the same population is one of the fundamental issues in statistics. There are dozens of tests that you can choose from to determine that "they are the same". You can use tests to compare means, medians, variances, rank-sums, and more. However, you HAVE to specify a test because each test has a different power.