[R-meta] metafor - within study non-independence

Jarrett Byrnes redbe@rd @ending from @rrr@net
Mon Aug 6 15:27:33 CEST 2018

Yes, and I’ve got some sample code if you need it. rma.mv is the way to go, although it’s unfortunately not as easy to put together a model with the structure you’re thinking of than lme4 or brms. 

For example, I have a study looking at fish response to kelp loss where we were examining how response varied by body size.

rma.mv(Hedges.G ~ Body.Size, VHedges.G,
       random=list(~1|Authors..Year., ~Authors..Year.|Fish.Species), 

Def look at ?rma.mv for information on struct as well as 
http://www.metafor-project.org/doku.php/tips:rma_vs_lm_lme_lmer <http://www.metafor-project.org/doku.php/tips:rma_vs_lm_lme_lmer>
http://www.metafor-project.org/doku.php/tips:rma.uni_vs_rma.mv <http://www.metafor-project.org/doku.php/tips:rma.uni_vs_rma.mv>
http://www.metafor-project.org/doku.php/analyses:konstantopoulos2011 <http://www.metafor-project.org/doku.php/analyses:konstantopoulos2011>
http://www.metafor-project.org/doku.php/tips:i2_multilevel_multivariate <http://www.metafor-project.org/doku.php/tips:i2_multilevel_multivariate>

There’s also the robumeta package from Elizabeth Tipton’s work.I also have some (very very very out of date) notes on this at http://jarrettbyrnes.info/meta/ <http://jarrettbyrnes.info/meta/> under April 30th.

Note, if it’s getting hairy, I’m also finding brms to be pretty good for my multilevel meta-analytic needs, but you’ve gotta go Bayesian.


Jarrett Byrnes
Assistant Professor
Department of Biology
University of Massachusetts Boston
100 Morrissey Blvd.
Boston, MA 02125

http://byrneslab.net <http://byrneslab.net/>
b: http://imachordata.com <http://imachordata.com/>
t: @jebyrnes
g+: http://gplus.to/jebyrnes <http://gplus.to/jebyrnes>
> On Aug 6, 2018, at 9:15 AM, Brian Cheng <bscheng using gmail.com> wrote:
> Dear list,
> Within the metafor package, is there a way to account for within study
> non-independence?  As an example from Viechtbauer 2010 J. of Stat.
> Soft. ("dat.bcg") what if Aronson 1948  conducted multiple trials
> within the study?
> In my field of ecology, it is common to conduct experiments with
> multiple moderator variables but with replicates at the "trial" level.
> In this context, it would be useful to break out each trial with the
> various moderators for modeling. I was thinking that this could be
> done more flexibly with lme4 (glmer) and a trial level random effect
> but there may be issues with that approach.
> Apologies if this has been asked before and thanks for your time!
> All the best,
> Brian
> --
> Brian S. Cheng, Ph.D.
> Assistant Professor of Marine Ecology
> Department of Environmental Conservation | University of Massachusetts Amherst
> (413) 545-2454 | bscheng using umass.edu | http://bscheng.com
> _______________________________________________
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis

	[[alternative HTML version deleted]]

More information about the R-sig-meta-analysis mailing list