Experimental design and analysis for variety trials
The construction of experimental designs has undergone considerable change in the last 25 years due to the power and availability of computers. Previously the construction of incomplete block designs was based upon their ease of analysis rather than on optimality. Nowadays, we have much more flexibility in the construction of designs since analysis is, in general, not a limiting factor. Extra blocking structures, such as row-column designs and latinized designs, can be incorporated into experimental designs to control field variation more effectively. Several examples are presented. Analysis of these designs can be carried out with recovery of inter-block information using residual maximum likelihood (REML), available in the latest versions of the major statistical packages such as GENSTAT. The use of neighbour methods to adjust for spatial variability in one or two dimensions has also been developed in recent years. It is now possible to present a unified model that includes both the intra-block and spatial analysis as special cases. These methods are illustrated using four sets of data from the UK.