ABSTRACT
The absence of any restriction on the allocation of treatments to experimental units in the completely randomized design has obvious disadvantages. In a field experiment it would be possible, as a result of the randomization, for all plots with a particular treatment to be grouped together in one corner of the experiment. Given possible soil fertility trends, this might produce a most misleading result. Or, in an animal feeding trial, one diet might be allocated to all the heaviest animals, which again could produce a very false result. Thus, while in the long run the completely randomized design gives a fair comparison of treatments, a particular experiment might prove unfortunate. The logical way to overcome this possibility of an unfortunate design is to restrict the randomization in some way. This must be done in an objective manner. Having obtained a design we cannot then decide whether to accept it or try again, or even worse, adjust it. The remedy lies in grouping the units in such a way that units in the same group are as similar as possible, and then allocate at random each treatment to one unit in each of the groups. Such a scheme requires that each group contains as many units as there are treatments.
