Open wlangera opened 1 year ago
Hi Ward,
Thank you for this issue.
According to your brms code, you have uncorrelated group-level intercepts and period
slopes for species
(because of the ||
syntax; see the brms documentation and Table 2 in the lme4 vignette Fitting Linear Mixed-Effects Models Using lme4; beware that "uncorrelated" refers to zero correlation of the group-level intercept (regarded as a random variable) and the group-level period
slope (regarded as a random variable), so to zero correlation in the corresponding bivariate normal distribution).
You are right that this is not supported by projpred yet. Sorry :/
I'll leave this issue open because (i) we should throw a more informative error message in that case and (ii) as a reminder that this could be implemented in the future.
I forgot to answer two more specific questions:
can I specify my model in a different way such that I can use uncorrelated random variables?
Currently, I don't think this is possible.
I also wonder if it matters for projection prediction whether they are specified as correlated or not.
Yes, it does. But the reason why I added some more explanations concerning the ||
syntax above is that given these explanations, you might perhaps decide that you don't need the ||
syntax and instead rely on the ordinary |
syntax.
Thank you for your reply and extra information!
Hi
I want to fit a Poisson model with uncorrelated random variables and perform projection prediction. Is it true that this is not yet possible with projpred or can I specify my model in a different way such that I can use uncorrelated random variables? I also wonder if it matters for projection prediction whether they are specified as correlated or not.
Unexpected number of | characters in group terms. Please contact the package maintainer.
Created on 2023-07-24 with reprex v2.0.2
Kind regards, Ward