[R-meta] Overdispersion in data analyized by metafor. Does one need to correct, if so how can is this accomplished?
jsorkin at som.umaryland.edu
Tue Oct 31 00:48:45 CET 2017
I thank Wolfgang Viechtbauer for the work he has done, and the time he has devoted to the metafor package.
I hope you will allow me to ask what may seem like two simple questions. My questions, more completely described below are (1) Is it necessary to account for over-dispersion when using metafor? (2) If it is necessary, how does one do so. My data, and R code are attached to this email as a text document (as a .R file).
I have data describing two groups, one labeled "regional" and a second "general". Before I discovered the metafor package, I used glm with Poisson regression to determine if the event rates differed between my two groups. Poisson regression give a significant difference, but the ratio of the residual variance to the degrees of freedom is far from one. I re-ran the glm using quaisPoisson [family=quasipoisson(link=log)] and got a no significant difference between groups. I then ran a negative binomial analysis and to make sure the finding of excessive deviance was correct (deviance 124, DF=30, ratio apprxomatly 4).
After I discovered metafor, I re-ran the analyses and get a significant group effect. I think the over-dispersion I found in the Poission and negative binomial regressions may be causing an estimation problem for metafor. If over-dispersion is a problem, I don't know how I can have metafor account for the over-dispersion (as I did with the glm by using quasipoisson and negative binomial rather than poisson regression).
I thank you for your time and help.
John David Sorkin M.D., Ph.D.
Professor of Medicine
Chief, Biostatistics and Informatics
University of Maryland School of Medicine Division of Gerontology and Geriatric Medicine
Baltimore VA Medical Center
10 North Greene Street
Baltimore, MD 21201-1524
(Fax) 410-605-7913 (Please call phone number above prior to faxing)
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 4094 bytes
More information about the R-sig-meta-analysis