Mixed modeling vs subsetting?

Hi all,

I have a microbiome dataset with multiple fixed (microbiome type, treatment) and random effects (sampling site, sampleid) and was wondering what is the latest on using mixed modeling to analyze such data. I am seeing lots of nice examples and even r packages in the literature but it seems like there is no consensus on what is the safest route.

I am asking because I have largely grown used to simply subsetting data to analyze a certain factor such as looking at Treatment X effects only in the foregut and subsetting all other samples out (i.e. hindgut samples). But it seems to me that mixed modeling, where you incorporate all factors may be more robust?

I have seen that ANCOM 2.0 now supports adding in covariates but it does not report the relative contributions of all fixed and random effects.

Ren et al.'s Bayesian model, adjusting for zero inflation, and Graham et al.'s MIMIX bayesian model seem promising but just wanted to ask this forum to see what the consensus on mixed modeling vs. subsetting was. Or if I am even asking the right question here.

Cheers,
Sam

1 Like