[R-sig-ME] Patterned Variance-Covariance matrices for the random effect

[Douglas Bates]

On Tue, Aug 4, 2009 at 2:22 PM, Tahira Jamil<tahirajamil at yahoo.com> wrote:
> Hi everyone

> I am wondering How can I assign special patteren / form e.g. compound symmetry structure , block diagonal etc to variance-covariace matrices to mer calss.

> As in nlme library we have pdMat classes to specify patterened variance-covariance matrices for the random effects.
> Hope to have some help from group members.

The functions in the lme4 package do not allow for arbitrary
specification of the variance-covariance matrix for the random
effects.  Generally you specify random effects terms in the model
formula and the variance-covariance matrix of the random effects is
derived from the rules

random effects have a multivariate Gaussian distribution with mean zero
random effects associated with different terms in the formula are independent
random effects associated with different levels of the grouping factor
in a given term are independent
random effects associated with the same level of the grouping factor
in a given term have a general positive semidefinite
variance-covariance structure that is common to all levels.