In which condition we apply 99% of CI. Which method is appropriate to increase number of sample size either through 99% of CI or going through design effect.
Dear Krishna,
I presume that you are asking about sample size calculations. The basic principle is that the smaller the effect you are seeking to observe or the greater the confidence at which it you wish to observed it then the more data you will need to detect it.
I believe you are asking a "which is better" question about whether to use confidence intervals or "design effect". I don't really understand exactly what you mean here. The only thing I can suggest is that it is unusual to use confidence intervals in a sample size calculation.
Again, a basic principle of confidence intervals is the higher the percentage confidence the wider the interval will be. This is slightly counterintuitive. Significant differences are associated with non-overlapping confidence intervals. Thus the wider the interval, the less likely they are not to overlap.
Mr. Peter rightly said that design effect and confidence interval are two different concepts. Generally we do not use confidence interval to calculate sample size. Design effect is used there to adjust the sample size if we are using any sampling method other than simple random sampling.
As you increase the confidence coefficient, the width of confidence interval will be increases.
Krishna -
I think you are confusing the terms "design effect," and "effect size." They are very different concepts that you might research.
The concern with estimating required sample size is that you first have to have bias under control, which is not generally true, but hopefully 'true enough,' and then you are trying to determine sample size, say for an estimated mean, by what is required with a given population standard deviation to arrive at a desired standard error (estimated) of the estimated mean. Whether you look at relative standard errors or correctly consider hypothesis tests (which are often incorrectly considered), you are generally looking to see what sample goes with what standard deviation to get what desired standard error. An effect size is just a difference between what you might have and what might be acceptable, but there are no absolutes here, so all has to be consider in terms of standard error, one way or another.
The situation is more complex when you stratify. Stratification of your sample and population is often useful, but now you have to consider a combination of these sample size, and population stratum standard deviations, and say mean standard errors, by stratum to get an overall standard error.
Cheers - Jim
PS - Confidence intervals, say of means, by the way, make use of standard errors.
It's all based on getting from a standard deviation to a standard error.
I mostly agree with Peter. However, I would particularly disagree with "... it is unusual to use confidence intervals in a sample size calculation." People often want a sample size that will give them a confidence interval of a given percent that is not unacceptably wide. - It all boils down to standard errors.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size