[R-meta] Dear Wolfgang

Thu, Jan 27, 2022 10:53 AM

Dear Juhyung,

Briefly, I am not aware of any proper evidence based guidelines for such a threshold. You sometimes see people mention that there should be 5 or 10 studies per moderator, but these are just heuristics. One could do a proper power analysis if one is worried about false negatives. See

https://www.jepusto.com/publication/power-approximations-for-dependent-effects/

for some recent work that might be applicable to your case (since your post implies that you are dealing with a more complex data structure).

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On
Behalf Of Lee, Ju
Sent: Wednesday, 26 January, 2022 23:07
To: r-sig-meta-analysis at r-project.org
Subject: [R-meta] Dear Wolfgang

Hello,

I am currently using a mixed effect meta-regression to explore the effects of
different environmental variables on fish densities in coastal habitats.
For this, I am constructing a different model for individual fish species, in
which my goal is to identify important predictor variables separately for each
species by using aicc-based model selection (glmulti).
For each fish dataset, I usually identify 3-4 potentially important predictor
variables ? which include both categorical and continuous variables ? based on a
priori hypothesis and the statistical test of omnibus test (of each individual
predictor variable).
I am only testing the main effects and not the interaction of multiple variables
being included in the final models.

The problem I am running into is the small number of studies being available for
many species, with the number of study response (effect size, not independent
study) ranging from 6 to 100 for different fish species.

So my question is:

Is there a commonly used threshold for the multiple meta-regression using rma.mv
to be reliable and avoid false positive or negative relationship?
From
https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0229345&ty
pe=printable I read that, for meta-regression, n should be greater than 8 for low
variance data, but should be greater than 25 for high variance data.
I wanted to seek your advice on what would be a good threshold criteria for
minimum study response (effect size) number for running the meta-regression
models.
Also, could we also apply similar threshold for number of independent studies
(not effect size, but the actual publication) included in each dataset as well?

Thank you.
Sincerely,
Juhyung

Juhyung Lee
Postdoctoral Fellow
Marine Science Center, Northeastern University
430 Nahant Rd, Nahant, MA 01908, USA
Phone: +1(650)285-7614

[R-meta] Dear Wolfgang

Thread (3 messages)