When doing a publication bias analysis and the standard errors (and the same sample sizes of different groups) is not readily available, is it possible to make the assumption that there were equal sample between the control and experimental group? So basically the total sample size divided by two and that number used to calculate the standard error which the publication bias is based on. If so, are there any papers that's done this before?
You would have to empirically test the equivalence assumption first, as is done in bioequivalence studies.
Why not just try to contact the authors and request this information?
Hmm...yes that does tend to happen, although it is more frequent if the paper is more than five-ten years old.
Anyway, I was thinking you could test whether an error in your assumption of equal Ns would actually have much of an effect. This could be done by calculating bias on the data you have, and then change N or the standard error, into, say, a ten percent difference between groups, recalculating and comparing with the original value.
You could also calculate how much sample sizes usually differ in the data you have. Given these two sources of information, you could estimate how much your assumption could possibly affect the results. Do you get it? This is just an idea of mine, I'm not sure what kind of test you're using, or how these tests usually use sample sizes, so I might be on the wrong track.
Its an interesting problem - I'll add it to my list of things to study.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis