My significant other, in getting his M.S. research ready for publication, introduced me to multilevel modeling, and we worked through a reanalysis together. Now, I’m also interested to apply this framework to my 2nd dissertation chapter before publishing, so I can model cohesive community response while still extracting species’ responses.
The following code block is copied straight from a forum post, but I wanted to save it in case the thread disappears someday. The author is converting multilevel models from lme to (Bayesian) R-INLA format.
“For each model I have first the lme formulation and then the formula for INLA” – Ignacio Quintero
# random intercept m1 = lmer(dr ~ st.ele + (1|poId), data = sdat) form = dr ~ st.ele + f(poId, model = 'iid') #organize data n.loc = max(sdat$poId) sdat$i.int = sdat$poId sdat$j.int = sdat$poId + n.loc sdat$k.int = sdat$j.int + n.loc # uncorrelated random intercept and random slope m2 = lmer(dr ~ st.ele + (1|poId) + (0 + st.ele|poId), data = sdat) form = dr ~ st.ele + f(i.int, model = 'iid') + f(j.int, st.ele, model = 'iid') # correlated random intercept and random slope m3 = lmer(dr ~ st.ele + (st.ele |poId), data = sdat) form = dr ~ st.ele + f(i.int, model = 'iid2d', n = 2*n.loc) + f(j.int, st.ele, copy = 'i.int') # uncorrelated random intercept and random slope of st.ele^2 m4 = lmer(dr ~ st.ele + I(st.ele^2) + (1|poId) + (0 + I(st.ele^2)|poId), data = sdat) form = dr ~ st.ele + I(st.ele^2) + f(i.int, model = 'iid') + f(j.int, I(st.ele^2), model = 'iid')
I’m currently running a model selection per community, to determine which variables are most important across each avian community.