Skip to content
Prev 19203 / 20628 Next

Comparison between the ouputs of the lmer and the brm (from the brms package) functions

sample_prior generates extra return values, it doesn't block sampling
from the posterior.

Thierry is right though -- you have < 100 observation and only 4 levels
of your grouping variable. Recall that random effects are *variance*
components -- estimating the variance with only 4 observations won't
give you great estimates. The maximum-likelihood estimate is well
defined, but the likelihood surface is probably quite flat and so you
can have a "highest point" on it, but there will be a lot of points that
are as nearly as high. And that's what the Bayesian analysis catches
better because it's not using a point estimate of the variance components.

Phillip
On 16/4/21 12:57 pm, Torsten Hauffe wrote: