I am learning multilevel meta-analysis (or, three-level meta-analysis e.g., https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/mlma.html, and Three-Level Multilevel Meta-analysis: What exactly are the three levels? Multilevel vs. Multilevel SEM Approaches) and trying to deeply understand how this new relatively new approach fits into the more commonly used general meta-analysis (or two-level meta-analyses) in relation to weighting.

For the general meta-analysis model, I learned that there are two different models based on the effect size weighting: fixed-effects model and random-effects model, the former uses the inverse variance weighting of sampling variance and the latter uses the inverse variance weighting of sampling variance plus tau^2 (i.e., between-study variances).

When talking about multilevel meta-analysis models, I've personally never heard that people discuss a model being fixed-effects or random-effects. I realized that metafor does not allow us to specify a fixed-effects model when running multilevel meta-analysis (although we can choose ML or REML for method). So, I am wondering if multilevel meta-analysis is theoretically a random-effects model?

So, my question is, Can we also call multilevel meta-analysis models either (i) fixed-effects model or (ii) random-effects model, based on the approach of weighting?

I would appreciate it if I could hear any perspectives, or where I can further read into, and parts that I am not understanding correctly if any.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.