I’m probably a newby but I just found out about Marginalized Multilevel Models (MMM) for clustered data.
So we have, conditional (random effects), marginal (GEE), and marginalized models.
Since MMM looks very nice from a theoretical point of view estimating a pop-level effect. From what I understand while there is ongoing debate on whether to use marginal or conditional approaches, MMMs demonstrate that latent structures can be used for both purposes. Marginalized models are able to handle both time-dependent and time-independent covariates while also accounting for within-subject dependence and unobserved random effects. In contrast to GEE, the MMM framework utilizes likelihood-based estimation and inference, allowing for the quantification of between-cluster heterogeneity via the variance of random effects, obtaining cluster-specific predictions.
These facts make them really powerful for example for omics (single cell or spatial transcriptomics) where data are all correlated an clustered, but since it seems that no one is using them, I got suspicious that there’s something under the hood that I can’t grasp and makes them unusable.
Do someone have any experience or thoughts?