Programming the statistical procedures from SAS

Is a sample size of 5 enough to run a glimmix model?

Reply
Contributor
Posts: 52

Is a sample size of 5 enough to run a glimmix model?

Anyone have input on this?

Super User
Posts: 17,784

Re: Is a sample size of 5 enough to run a glimmix model?

Not in my opinion. 

 

 

Contributor
Posts: 52

Re: Is a sample size of 5 enough to run a glimmix model?

Would reviewers frown if results are reported without any formal statistical tests in such a case? I am wondering what is commonly done in such a scenario.

Super User
Posts: 17,784

Re: Is a sample size of 5 enough to run a glimmix model?

In my limited experience, they're not reported as statistical results, more of case overviews. The results cannot be generalized to any population so really, it's infomative depending on the subject but you can't really test anything with 5 numbers. It's more story telling and to be honest, not really worth a journal publication. Maybe a blog post but for n=5. This may vary a bit depending on the subject area, but if it's not a rare disease it's really hard to justify a publication on 5 people. For one, re-identification may be easy. 

 

Here's an example I once heard that I use to explain the issue of sample size. What's the odds of having all babies born at a hospital be female? If the hospital has 5 live births in a day? If the hospital has 20 live births in a day? If the hospital has 100 live births in a day?

Contributor
Posts: 52

Re: Is a sample size of 5 enough to run a glimmix model?

I see your point. I am a bit worried about the reviewers, I suppose it will depend on how statistically savvy they are. It is a very rare intervention, so i think it is publication worthy.

Trusted Advisor
Posts: 1,610

Re: Is a sample size of 5 enough to run a glimmix model?


Melk wrote:

I see your point. I am a bit worried about the reviewers, I suppose it will depend on how statistically savvy they are. It is a very rare intervention, so i think it is publication worthy.


And even if it is a very rare intervention, this points out one of the benefits to society of doing the proper statistical testing ... it would prevent someone from claiming he had solved a problem with a very small amount of data. On the other hand, it also points out one of the drawbacks to society ... a great invention may not be discovered (or the discovery delayed) without the proper amount of data.

 

As others have said, it depends on how you present this. If you state that this is an interesting result AND MORE STUDY IS NEEDED, I think maybe you have a chance.

Super User
Posts: 17,784

Re: Is a sample size of 5 enough to run a glimmix model?


PaigeMiller wrote:

 

As others have said, it depends on how you present this. If you state that this is an interesting result AND MORE STUDY IS NEEDED, I think maybe you have a chance.


X100!!!

 

Present as "this is interesting, shows promise and a good rationale for further research".

Contributor
Posts: 52

Re: Is a sample size of 5 enough to run a glimmix model?

Thanks everyone, will take a shot at it and will incorporate your comments in the summary of the findings!

Super User
Posts: 10,483

Re: Is a sample size of 5 enough to run a glimmix model?


Melk wrote:

Anyone have input on this?


About the only statistical test I would feel comfortable running with a sample of 5 would be a binomial test. Testing something like "Was the proportion of event x in the = p ?" One of the major problems will be that your test rejection areas are not going to be "nice". Look up references on exact binomial tests and you'll find out why.

Ask a Question
Discussion stats
  • 8 replies
  • 253 views
  • 7 likes
  • 4 in conversation