Choosing the Most Appropriate Experimental Design
This chapter reviews the important concepts of experimental design, placing the various designs considered in context and indicating where more complex designs that have not been discussed are appropriate. It emphasizes that the following two concepts are extremely important in designing experiments: control of variation by blocking and factorial structure for treatments. If the comparison of treatment and control is sufficiently important, then the design should be chosen, even though the precision for comparing treatments other than the control is clearly worse. It is essential to distinguish two components of an experimental design, the structure of the experimental units, and the structure of the treatments. The important ideas about units are those of replication, blocking, and randomization. There can be no independent randomization at the split-plot level, and this invalidates the randomization justification for split-plot comparisons. The special feature of split-plot designs is that different groups of treatments are deliberately applied to units of different sizes and levels of variation.