Skip to content
Prev 19920 / 20628 Next

Problem specifying uncorrelated random intercepts and slopes for a multi-df covariate

Including correlation parameters can "pull" random effects (so to speak) 
in the direction of the estimated correlation. See here:
https://doingbayesiandataanalysis.blogspot.com/2019/07/shrinkage-in-hierarchical-models-random.html

So perhaps the slopes are changing quite a bit due to their correlation 
with the intercept?
You could do some experimentation to test this, for example, keeping the 
correlation between slopes, but removing the intercept-slope correlations.
Maybe something like this would do it: (1 | id) + (0 + x2B + x2C | id)

(that said, it does seem a bit strange to me that both of them are zero 
when not estimating correlations)

Jo?o
On 05/08/2022 16:16, Juho Kristian Ruohonen wrote: