I’m wondering whether brms can be used for structural equation modeling (SEM) when combining models with different distributions? I see it’s possible to conduct multivariate models with multiple response variables (https://cran.r-project.org/web/packages/brms/vignettes/brms_multivariate.html), and this page uses the method for running simple SEM (https://rpubs.com/jebyrnes/brms_bayes_sem). However, this example contains models using only Gaussian distributions.
My model contains two components, a negative binomial model and an ordinal model. My understanding as the coefficients (estimates) returned are not comparable between the two models as they are unstandardized and provide different information. Is this true? If so, is there are way to standardize them so I can compare the relative importance of different paths across models?
I understand the ordinal model might be adding extra complications given it is categorical data. If I was able to re-model this as continuous data would this make running SEM easier (i.e. if I was trying to combine a negative binomial and Gaussian model)?
These are my models:
Seedlings.germinated ~ Fertiliser + Shelter + pasture.cover + (1|Block), family = negbinomial()
pasture.cover. ~ Fertiliser + Shelter + Grass.treatment + (1|Block), family = cumulative()
Any indication of whether this is possible or not would be greatly appreciated.
Operating system: Windows
brms version: 2.15.0