This chapter is primarily devoted to experiments that compare 2 treatments with respect to an outcome measure. Six design scenarios are discussed: (a) completely randomized designs (treatments are assigned completely at random); (b) randomized block designs (experimental units are subdivided into blocks of like subjects, with one subject in each block randomly assigned to each treatment); (c) stratified designs (subjects are categorized into subpopulations called strata, and within each stratum, a completely randomized design is conducted); (d) crossover designs (each subject gets both treatments, but order is completely at random); (e) 2 � 2 factorial designs [design can be in any of the formats (a)–(d) but there are 4 not 2 treatments representing 2 types of treatment interventions, each with 2 levels]; and (f) randomized designs with “random” effects. This is much like the stratified design, except there is only 1 sample, at least conceptually, from thestrata. Examples might be litters of laboratory animals, surgical practices, or batches of a therapeutic agent. The desire is to make inferences about treatments in the population as a whole, not just in the strata that were actually sampled.
