ABSTRACT
This chapter considers aspects of inference and design for a spatial linear model that are peculiar to spatial experiments. It begins by clarifying the connections between randomization, the classical linear model (which ignores spatial correlation), and the spatial linear model, and by describing how to use them properly. Next, optimal spatial design of experiments is considered. Simulations are used to illustrate that the design can have a large impact on the precision of estimated treatment effects and contrasts, with designs in which treatments are spatially well-interspersed among the experimental units generally performing best. Particular classes of designs that have this property, called nearest-neighbor balanced block designs and nearest-neighbor balanced designs, are featured. Finally, the methods of analysis and design are illustrated for the caribou forage experiment.
