Greetings,
I am Mamun. I just want to know that if I use 7% margin of error than will it cause any problem in future analysis? I am using stratified sampling technique, for this if i take 7% margin of error then i get small sample size. It would be better for me to take small sample size.
Thank You.
In stratified sampling your are going to sample two heterogenous group. In this case 7 percent level of of procession is not good because you are treating the population as more of homogeneous. Hence,it is better to take 5 percent.
It is prefered to be 1-5%. 7% margin of error may be acceptable but it depends on the research objectives, if it is very precise or more exploratory. If limited resources are available and for that a small sample size is conducted. If the population is relatively homogeneous or the characteristics of interest are consistent across strata, a 7% margin of error may provide adequate precision.
Mamun Ur Rashid -
Remember that your accuracy can easily be worse than that 7%. You have to ask yourself if, in your application, would an 'error' that large or somewhat larger be meaningful for you? If not, and it does not help, then this may be worse than useless in that results could be misleading.
Also, remember that your overall result is what you are talking about with a stratified random sample (or a stratified prediction-based approach). So results for a given strata are even less accurate. I think people seem to sometimes forget that if they want results for individual strata, they need larger sample sizes and to treat each stratum as a subpopulation, and thus you do not really have a stratified sample, but samples from each subpopulation. I don't know that you are doing that, but I thought I'd mention that problem.
Cheers - Jim
I concur with all previous comments.
But as an agnostic ex-econometrician, I will highlight Heba Ramadan 's one.
If you are dealing with matters of life and death in two different populations, 0.05 is the bare minimum, with a robust sample size for both strata. And it will be just the beginning of a long trip: you will have to look at the effect size to check if you are on a solid statistical ground for your test, and check for any issue in your model specification.
Usually, it is a very depressing exercise as most research results don't survive.
But if your interest is more exploratory, then 10% is more than enough as long as you don't trust your results but use them as clues rather than proofs.
Even if it is a matter of life and death and if you have to make a decision at the end of your research, odds ratio will guide you, even if you won't be certain to make the right decision.
My two cents,
Hubert
Just to be clear, Mamun Ur Rashid, I think you are not talking about a p-value, but the margin of error you are considering. Is that correct?
James R Knaub Yes, it's correct.
Thank you very much for your answer. It's really helpful for me.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Control Theory
- Control Systems