ROBUST LINEAR BLOCK-DESIGNS FOR A SUSPECTED LV MODEL

Research on optimal design for dependent observations has so far concentrated on the situation where the dependence structure has a specified parametric form. In many cases the results depend on the actual value of the parameter. However, the form of the dependence cannot usually be specified before the experiment is carried out. In this paper we try to find designs that are efficient over a range of dependence structures which are plausible for a given situation. The linear variance (LV) model has been considered by many authors to be widely applicable to block experiments in which the plots have a one-dimensional layout. We seek designs that are efficient over a range of parameter values that are considered reasonable for the LV model, and which are also efficient for some reasonable alternative models.