by Schmelter, T..
Series: 2005-02, Preprints
In this paper the class of designs for linear mixed models is considered where each individual is observed under a (possibly different) approximate design and the number of observations is the same for all individuals. It is then shown that the $\Phi$-optimal design in the class of balanced designs (single-group designs) for estimating the population parameters is also optimal in that wider class of designs mentioned before, where $\Phi$ is a convex and monotone criterion function that has to satisfy mild assumptions, which is e.g. the case for D-optimality and all linear criteria.
optimal design, mixed model, random coefficient regresion