Optimal designs for minimax-criteria in random coefficient regression models (Q2423172): Difference between revisions

The author considers random coefficient regression models for \(n\) individuals with \(m\) observations per individual and looks for optimal designs for the best linear unbiased prediction of the individual response curve. As basic criterion for prediction the integrated mean squared error (IMSE) is used. The individual random regression parameters are assumed to have correlated components with known covariance matrix \(D\). Finding optimal designs the author focuses on a minimax criterion which minimizes the ``worst case'' for IMSE with respect to some variances in \(D\). Assuming a diagonal matrix \(D\) with small variances optimal designs are obtained for straight line and quadratic regression.