[R-meta] weighting a multilevel model and quality effects model

[Diego Grados Bedoya]

Hi Wolfgang,

Thank you for the advice and the reference.

Greetings,

Diego

On Sat, 31 Jul 2021 at 11:41, Wolfgang Viechtbauer <
wolfgang.viechtbauer using maastrichtuniversity.nl> wrote:

> Hi Diego,
>
> Not a direct response to your questions, but I would suggest to take a
> look at:
>
> https://www.metafor-project.org/doku.php/tips:weights_in_rma.mv_models
>
> The weighting that happens in such models (automatically) is more complex,
> so unless you have a very thorough understanding of this, I would *not*
> recommend to try to use the 'W' argument.
>
> Best,
> Wolfgang
>
> --
> Wolfgang Viechtbauer, Ph.D., Statistician | Department of Psychiatry and
> Neuropsychology | Maastricht University | P.O. Box 616 (VIJV1) | 6200 MD
> Maastricht, The Netherlands | +31 (43) 3884170 | https://www.wvbauer.com
>
> On Fri, 30 Jul 2021, Diego Grados Bedoya wrote:
>
> > Dear all,
> >
> > I am fitting the following multilevel model using the metafor package:
> >
> > model_0 <- rma.mv(yi = y, V = var.y, mod = ~ mod.y - 1, random = ~ 1 |
> > study.id / es.id, data = dat, method = "REML")
> >
> > I would like to explore the incorporation of weighting factors such as:
> >
> > model_1 <- rma.mv(yi = y, V = var.y, W = 1 / var.y, mod = ~ mod.y - 1,
> > random = ~ 1 | study.id / es.id, data = dat, method = "REML")
> > model_2 <- rma.mv(yi = y, V = var.y, W = sample.size.y, mod = ~ mod.y -
> 1,
> > random = ~ 1 | study.id / es.id, data = dat, method = "REML")
> >
> > Also assessing a quality effect model according to
> > http://dx.doi.org/10.1016/j.cct.2015.05.010, which defines a model
> based on
> > a relative quality index for each individual study (quality.y) varying
> from
> > 0-1:
> >
> > model_3 <- rma.mv(yi = y, V = var.y, W = quality.y / var.y, mod = ~
> mod.y -
> > 1, random = ~ 1 | study.id / es.id, data = dat, method = "REML")
> >
> > I have a couple of questions:
> >
> > 1) What is the rationality of incorporating weighting factors in the
> > multilevel model?. For instance, model_1 is giving more weight to more
> > precise studies/effect sizes while model_2 is giving more weight to big
> > studies/effect sizes ...
> > 2) Is the implementation of model_3 correct? Is there any other
> > alternative/method to incorporate a quality index?
> >
> > Any hint or literature is more than welcome,
> > Thank you in advance,
> >
> > Diego
>

	[[alternative HTML version deleted]]