This question already has an answer here:

I am modeling wood properties variation of several individual within the same tree species using multilevel lme(). I specified a first model as :

full <- lme(id ~ dist2 + I(dist2^2)
                 ,random=list(ind=pdDiag(~ dist2 + I(dist2^2)),
                              radius=pdDiag(~ dist2 + I(dist2^2)))
                 ,control=ctrl, meth='ML')

with 'radius' nested within 'ind'.

As 'id' is a longitudinal variable according to 'dist2' within 'radius' within 'ind', I want to add an AR(1) autocorrelation error term.

So, I wrote a second model by updating the first one with :

fullc <-   update(full,  correlation=corARMA(p=1,q=0, form = ~ 1 | ind/radius))

Is it a good way to include autocorrelation error structure in my model?

This specification is also possible and gives slightly different results :

fullc <-   update(full,  correlation=corARMA(p=1,q=0, form = ~ dist2 | ind/radius))

But I don't clearly understand the difference between both specifications. Could you help me ?

Thank you.



marked as duplicate by gung - Reinstate Monica r Aug 10 '18 at 0:35

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.