[R-meta] robust variance estimation with few clusters

[James Pustejovsky]

Yes, I think the same reasoning extends to meta-regression models as well.

On Thu, Jul 25, 2024 at 11:17 AM Filippo Gambarota <
filippo.gambarota using gmail.com> wrote:

> thank you James, I guess that the same reasoning could be applied to
> meta-regression right? I can just ignore the dependency and then apply the
> cluster robust function
>
> On Thu, 25 Jul 2024 at 16:42, James Pustejovsky <jepusto using gmail.com> wrote:
>
>> Hi Filippo,
>>
>> Your approach is to fit a univariate model and then cluster the standard
>> errors to allow for dependence across effects nested within papers. RVE is
>> based on using working models for the dependence structure, and so we might
>> call this approach an "independent effects" working model. Your approach
>> seems totally reasonable to me, especially if the higher-level structure is
>> difficult to estimate (because of few effects per cluster).
>>
>> James
>>
>> On Thu, Jul 25, 2024 at 8:08 AM Filippo Gambarota via R-sig-meta-analysis
>> <r-sig-meta-analysis using r-project.org> wrote:
>>
>>> Hi,
>>> I'm approaching the robust variance estimation using clubsandwhich and
>>> following the metafor documentation, however I'm not sure about a point.
>>> I
>>> have a dataset with a multilevel structure (independent effects within a
>>> paper). However the number of clusters is low as well as the number of
>>> effects within clusters.
>>>
>>> My idea was to fit a standard two-level model ignoring the multilevel
>>> structure (author/id is the paper/effect structure):
>>>
>>> ```
>>> fit <- rma(yi, vi, data = data)
>>> robust(fit, cluster = author, clubSandwich = TRUE)
>>> ```
>>> However in the metafor documentation the robust function is used on a
>>> multilevel model. Do I have to fit a multilevel model (with the problem
>>> of
>>> having few clusters) and then the robust function or my approach is
>>> correct?
>>>
>>> thank you
>>>
>>> --
>>> *Filippo Gambarota, PhD*
>>> Postdoctoral Researcher - University of Padova
>>> Department of Developmental and Social Psychology
>>> Website: filippogambarota.xyz
>>> Research Groups: Colab <http://colab.psy.unipd.it/>   Psicostat
>>> <https://psicostat.dpss.psy.unipd.it/>
>>>
>>>         [[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> 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
>>>
>>
>
> --
> *Filippo Gambarota, PhD*
> Postdoctoral Researcher - University of Padova
> Department of Developmental and Social Psychology
> Website: filippogambarota.xyz
> Research Groups: Colab <http://colab.psy.unipd.it/>   Psicostat
> <https://psicostat.dpss.psy.unipd.it/>
>

	[[alternative HTML version deleted]]