I like to estimate a mixed model with two Random Effects, that are independent of each other and among themselves.
I use Panel data with a nested structure (counties $j$ nested within regions $i$). The error component should decompose in the following way:
$$E_{ijt} = \alpha_i + \mu_{ij} + e_{ijt}$$
$\alpha_i$ is an unobservable region specific time-invariant effect (normal distributed)
$\mu_{ij}$ is a nested effect of county $j$ within region $i$ (normal distributed)
$e_{ijt}$ is a remainder disturbance (normal distributed)
I read somewhere that "lmer" is capable to fit such specifications with idenpendent Random Effects.
lmer(y~.... + (1|region) + (1|county)
Is the lmer-model adequate for my task?
Best Answer
Yes,
lme4
should be fine in here. The formula you are asking is:y ~ 1 + (1|region) + (1|region:county)
because in model you described
country
is nested inregion
.For an extended description and examples see here, here or here.