Dependency structure

Just to clarify, you mean that you want to specify a correlation structure
between the individual level error term in the model (also called the
residuals) and the random intercept or group-level error.

This doesn't make a lot of sense to me because the random intercept is
literally the product of a decomposition of the general model's error
structure into the within group (R matrix) and between group (G matrix)
components of the error. They are uncorrelated by construction. The only
way that they could possibly be correlated would be if you had an
exchangability problem in the random effects structure. You could have a
fuzzy boundaries issue like US counties are correlated by space. But you
wouldn't solve that by correlating the lower level error term with the
random intercept. You'd build a group boundary spatial weights matrix and
include it in the model.

I must be missing something in the translation.

Dependency structure

Thread (3 messages)