I performed a meta-analysis using 'metafor' and used rma.mv() to explore the effects of moderator variables on observed effect size. However, there seems to be no tool for model selection in 'metafor', and 'MuMIn' also cannot handle the objects fitted by rma. Anybody knows how to do model selection for meta-analysis models? Thanks ahead for your attention.
You should get information criteria out. That should allow you to manually calculate Aikaike weights etc. if that's of use. For example see:
Article AIC model selection using Akaike weights
There are tools for this in R for some models but I'm not sure about metafor.
Dear Thom, thanks for your kind suggestion. Actually, rma gives AIC value for each model. The problem is, it would be very time costly to manually build all the potential candidate models one by one, when number of explanatory factors are relatively large (for example, 5 or 6), which is common in ecological studies.
Best wishes, Ling-Ying
Here's an approach that might fit your needs:
https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/metareg.html
See section 8.3.3.5 Multi-Model Inference; this will show you some convenient ways to choose predictor combinations that will result in the best fit (e.g., based on AICc). Note that this uses relies on the 'dmetar' package which should be compatible with your output from metafor.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability