How many = a survey?

[Sheiila]

What the heck, I'll throw this question out to the group (you're all veyr helpful!)... I am going to do some research, about family docs and pediatricians' preparedness to recognize and diagnose a particular illness. How many doctors should I survey, in order for this data to be considered legit to a medical journal? Any idea?

---

Members don't see this ad.

 

[Djanaba]

That would depend on what sort of statistical analysis you intend to do with your data. Sample size and power of a test are directly related: more is usually better, especially if differences are subtle. If you're really unsure, talk with a researcher or stats person who can help you determine what you want your survey to find out, how to analyze your data, and from that pull out what sorts of results you think you'll see. This should give you a better idea if it should be 10, 100, or 1000. (But I can tell you already it's not going to be 10, and good luck getting a decent response rate: doctors are notoriously bad survey responders.)

If you're doing sponsored research in conjunction with your school, talk with your advisor. You'll need the affiliation for publication anyways, and likely a run through a human subjects committee.

 

[MikeS 78]

it'll be awfully hard to get a clinical chart review/series type evaluation into a refereed journal without a sponsor with an MD.....unless of course you are a 9 year old girl doing a science fair project on alternative medicine.....then people are beating down your door

 

[deziballer]

that depends on your study design. If you decide to do a cohort study, then your'e probably gonna need a very large sample size (couple hundred to thousands I'd guess). On the other hand, if you go for a case-control study (in this case, instead of comparing controls and cases, your'e studying different kinds of dcotrs), a sample of even 20 from each of the 2 specialties would make for a valid study. There's a bunch of other study designs which I'm not too familiar with, but you could find in an epidemiology textbook.
hope that helps, and good luck with your study.

 

[zolie]

There is a program called Epi-info that helps you to calculate samples sizes for given study types. In biostatistics an n above thirty is considered a "normal" sample, but in my experience I've found that studies almost always require at least 100 sample point. Of course, as someone pointed out above, it depends on the type of study you design etc.

 

[kwilson]

You can get the Epi Info software free from the CDC website (www.cdc.gov). You have to know about the type of study design, and what your hypothesis is (ie which group will recognize and diagnose more, and what the expected difference is). The smaller the difference you want to be able to detect, the larger sample size you'll need. If this is a mailed survey, expect about a 40-60% response rate, and increase your sample accordingly. Also, if you plan any multivariate modeling the standard rule of thumb is to have at least 50 subjects; you'll probably want more. The important thing for the journal is being able to show a study design consistent with your hypothesis. Depending on where you are submitting, having a coauthor who has published may be helpful. Also, if you work at a med center currently you may have to submit the proposal to a human subjects review committee (some journals request evidence of human subjects protection, too). What do you want to study?

Good luck! Research is fun