[R-sig-ME] Kronecker Product Covariance matrix in nlme?

[Mareike Kohlmann]

Hi,

I want to fit a multivariate covariance pattern model by assuming a
Kronecker product structure kronecker(U,V) for the covariance matrix
(Dimension pq x pq). This can be applied to multivariate longitudinal
data, for example, where the matrix U (with dimension qxq) is the
covariance matrix for the variables and V (with dimension p x p) the
covariance matrix to model the time dependencies.

In SAS proc mixed, this can be achieved by specifying UN at CS, UN at AR1 or
UN at UN (UN for the variables/ CS, AR1 or UN for the time structure) in the
REPEATED statement.

Is is possible to implement this model in nlme?

I suppose I have to write my own "corStruct" class and then use the weight
statement to get variable-specific variance estimates?

Thanks for your help!

Mareike


***************************************
Mareike Kohlmann
Department of Statistics
Ludwig-Maximilians-University Munich
Germany
E-Mail: mareike.kohlmann at stat.uni-muenchen.de