Locally \(D\)-optimal designs for heteroscedastic polynomial measurement error models (Q2189755)

The authors extend the work of \textit{M. Konstantinou} and \textit{H. Dette} [Biometrika 102, No. 4, 951--958 (2015; Zbl 1372.62029)] to heteroscedastic cases. They first develop the necessary optimality condition for Kiefer's \(\Phi_p\)-criteria (\(p \in[-\infty,1)\)) based on the correction approach for the weighted polynomial measurement error model, and then they present the upper bounds of the number of support points of locally \(D\)-optimal designs. For the simple linear measurement error model the locally \(D\)-optimal design can be constructed explicitely. The numerical results show that locally \(D\)-optimal designs are sensitive wrt heteroscedastic assumptions and model parameters.