ABSTRACT
This chapter describes a framework for balancing the mildly competing considerations in designing for efficient parameter estimation and for efficient prediction. Sampling designs that are efficient for parameter estimation may be inefficient for prediction, and vice versa. A combination of theoretical and empirical evidence suggests that completely regular designs generally lead to efficient spatial prediction provided model parameters are known. The planning, or design, of any scientific investigation involves many considerations, most obviously in the current context the number of measurement locations. An inhibitory design represents a compromise between a completely random and a completely regular design. As with non-adaptive designs, the choice of performance criterion is of paramount importance. The most important theoretical consideration in constructing an optimal non-adaptive design is whether the assumed model is at least approximately correct.
