[R-meta] unweighted analysis questions
Dear Alan
On 21/03/2024 14:51, Alan Wilson via R-sig-meta-analysis wrote:
Hey everyone - I have two questions regarding unweighted analyses. 1) Using the suggested code for unweighted analysis (rma(yi, vi, method="FE", data=dat, weighted=FALSE)) provides the same main effect but different 95% CIs when using the same or different vi values across studies. I also found different 95% CIs when testing two different vi values (1 vs 100) that were the same across studies. I also found that changing weighted=FALSE to weighted=TRUE provided same 95% CIs when using the same dataset. Any idea what is causing the variation in 95% CIs? What is best for unweighted analyses?
That is what I would have expected. Although the vi are not being used as weights they do still tell you about the variability. If you combine lots of very imprecise studies you would not expect to get the same overall CI as combining lots of very precise studies. Otherwise you would be getting a free lunch.
2) Any suggested approaches for publication bias checks for unweighted analyses? I was thinking funnel plots of effect size and sample size.
Other people will probably pitch in with evidence-based ideas here but I would go for sqrt sample size I think. Michael
Thanks for any advice. alan [[alternative HTML version deleted]]
_______________________________________________ R-sig-meta-analysis mailing list @ R-sig-meta-analysis at r-project.org To manage your subscription to this mailing list, go to: https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
Michael