Standard Errors for EM Estimates in Variance Component Models

A procedure is derived for computing standard errors in random
intercept models for estimates obtained from the EM algorithm. We
discuss two different approaches: a Gauss-Hermite quadrature for
Gaussian random effect models and a nonparametric maximum
likelihood estimation for an unspecified random effect
distribution. An approximation of the expected Fisher information
matrix is proposed which is based on an expansion of the EM
estimating equation. This allows for inferential arguments based on
EM estimates, as demonstrated by an example and simulations.