Criterion-robust designs for the models of spring balance weighing (Q2913235)

The paper presents a study of the experimental designs which have the maximum minimal efficiency, in the class of all orthogonally invariant information functions, for the linear regression model of the first degree on the vertices of a \(d\)-dimensional unit cube. This model has a nice interpretation related to the weighings in a spring balance, and the authors also study the effect of an intercept term which indicates the relative importance of a bias in the balance.NEWLINENEWLINEThis paper applies some theoretical results of previous work to give an analytic formula for the maximin efficient design in the case without bias, and successfully applies the semidefinite programming method described in a previous paper to compute the maximin efficient design in presence of an intercept parameter.NEWLINENEWLINEThe criterion of minimal efficiency is probably one of the most complicated in the optimal experimental design theory, since it relies on complicated non-differentiable functions. However, it seems very natural to use a design maximizing the min-efficiency for reasons of robustness. Therefore, it is very satisfactory (and somehow unexpected) to see that there is an analytic formula for the maximim efficient design in the model of spring balance weighing. The application of semidefinite programming for the case of a biased balance is also very interesting.