I recently noticed that depending on the order I write the predictors for in a lasso based mixed model as implemented in glmmlasso in R documented here
I get different coefficients, z, and p values which complicates interpretation. I am wondering if there is an explanation on why this would be the case (or if it should even be the case). I saw a related question based on glmnethere which also does some lasso regularisation and based on answers to a similar issue on that package and it seems that this shouldn't happen.
one example of the syntax for such moel is
lm2 <- glmmLasso(dv ~ v1+ v2+v3+v4+v5+v6,
rnd = list(v7 =~1, v8=~1),
lambda=10,
data = data, final.re=TRUE)
Appreciate any pointers
