I am exploring hierarchical logistic regression, using glmer from the lme4 package. To my understanding, one of the first steps in multilevel modeling is to estimate the degree of clustering of level-1 units within level-2 units, given by the intraclass correlation (to "justify" the additional cost of estimating parameters to account for the clustering). When I run a fully unconditional model with glmer
fitMLnull <- glmer(outcome ~ 1 + (1|level2.ID), family=binomial)
the glmer fit gives me the variance in the intercept for outcome at level-2, but the residual level-1 variance in outcome is nowhere to be found. I read somewhere (can't track it down now) that the residual level-1 variance is not estimated in HGLM (or at least in glmer). Is this true? If so, is there an alternative way to approximate the degree of clustering in the data? If not, how can I access this value to calculate the ICC?