[R-meta] Selection models for rma.mv-objects

[James Pustejovsky]

We only looked at a selection model (specifically, the three-parameter
selection model) when applied to aggregated effect sizes or when ignoring
the issue of dependent effects.

I agree with Wolfgang's assessment that multivariate selection models
remain an open problem.

One other possibility: Mathur and Vanderweele (2020;
https://doi.org/10.1111/rssc.12440) have a sensitivity analysis approach
that is conceptually similar to the three-parameter selection model and
that works with dependent effect sizes (via robust variance estimation).
It's implemented in the PublicationBias R package:
https://cran.r-project.org/web/packages/PublicationBias/index.html

James

On Mon, Sep 19, 2022 at 3:59 AM Viechtbauer, Wolfgang (NP) <
wolfgang.viechtbauer using maastrichtuniversity.nl> wrote:

> Dear Therese,
>
> Implementing selection models for multilevel/multivariate models is very
> difficult, because it might involve the repeated computation of multiple
> integrals (while the selection models for 'rma.uni' objects involve
> repeated computation of simple integrals, which is already numerically
> challenging). Hence, as far as I know, nobody has implemented selection
> models for multilevel/multivariate models.
>
> This aside, for multilevel/multivariate models, one also has to carefully
> consider where/how selection may have taken place. For the individual
> effect sizes? Or for entire studies altogether? Or a combination of the
> two? For modeling selection of entire studies, the question is also what
> the selection depends on. The p-value of the most significant effect size?
> Or the p-value of what might have been the focal effect size within a
> study? Or the p-value of some kind of omnibus test of multiple effect sizes?
>
> At the moment, I would say the regression test is the most straightforward
> technique that generalizes and that one could use for
> multilevel/multivariate models. See also:
>
> https://doi.org/10.1037/met0000300
>
> The paper also examines a selection model, but it wasn't clear to me
> whether this was some kind of multivariate version of it or not. Maybe
> James (or Melissa - but I am not sure if she is subscribed to this mailing
> list) can comment on this.
>
> Best,
> Wolfgang
>
> >-----Original Message-----
> >From: R-sig-meta-analysis [mailto:
> r-sig-meta-analysis-bounces using r-project.org] On
> >Behalf Of Thölking, Theresa
> >Sent: Monday, 19 September, 2022 10:22
> >To: r-sig-meta-analysis using r-project.org
> >Subject: [R-meta] Selection models for rma.mv-objects
> >
> >Hello everybody,
> >
> >is there a function available in R that enables selection models for
> rma.mv-
> >objects? I could only find an implementation for rma.uni.
> >
> >Thanks in advance.
> >
> >Best,
> >Theresa
>
> _______________________________________________
> R-sig-meta-analysis mailing list @ R-sig-meta-analysis using r-project.org
> To manage your subscription to this mailing list, go to:
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
>

	[[alternative HTML version deleted]]