A randomization problem in forming designs with superimposed treatments

Some programmes of experimentation require the superimposition of a new set of treatments on an existing experiment. Randomization of such treatments has received little attention in the literature but merits careful consideration. The present paper examines two apparently different theoretical approaches, one in effect randomizing the experiments together and the other randomizing the second experiment without reference to any randomization used in the first. In the simplest superimpositions the two methods give identical results in practice but problems arise when one Latin square is to be superimposed on another.