[R-meta] Meta-analysis including multiple intervention arms
|o@n@@@||n@@cr|@te@ @end|ng |rom gm@||@com
Sun Nov 15 16:31:40 CET 2020
Thank you both. Yes, to clarify I am using correlated effects, and the
study with more effects is only considered once.
See here: https://cran.r-project.org/web/packages/robumeta/robumeta.pdf
On Sun, Nov 15, 2020 at 4:14 PM James Pustejovsky <jepusto using gmail.com> wrote:
> It depends on the structure of the random effects (i.e., the “working
> model” in the RVE framework).
> If one used a model with independent random effects for each ES estimate,
> then yes, increased heterogeneity would mean that the study with 2 effects
> receives more weight.
> But the “correlated effects” working model has a study-level random
> effect, which hits all ES estimates within the study. In the case, the
> study ends up getting the appropriate amount of weight in the overall
> > On Nov 15, 2020, at 8:34 AM, Guido Schwarzer <sc using imbi.uni-freiburg.de>
> > Am 14.11.20 um 16:29 schrieb James Pustejovsky:
> >> [...] Splitting the control sample size would over-correct, so that the
> study would end up with less weight than it should receive.
> > Is this also true if you have a (very) large between-study variance?
> > In such cases each estimate gets essentially the same weight in a
> meta-analysis which means that a study with two estimates gets twice the
> weight (not sure about this relationsship if using RVE).
> > Best wishes
> > Guido
Ioana-Alina Cristea, Ph.D.
Department of Brain and Behavioral Sciences
University of Pavia, Piazza Botta 11, 27100 Pavia, Italy
METRICS (Meta-research Innovation Center at Stanford)
Stanford University, California, USA
Editorial Advisory Board Lancet Psychiatry
International Editorial Board BJPsych Open
[[alternative HTML version deleted]]
More information about the R-sig-meta-analysis