A procedure for allocating heterogenous units to treatment groups

Nokoe, S. & Zewdie, S. (2000). A procedure for allocating heterogeneous units to treatment groups. Discovery and Innovation, 12(3/4), 179-198.

In allocating experimental subjects that are not homogeneous (eg. animals, socio-economic units such as households, etc.) to treatment groups, certain attributes which are likely to affect the results if not accounted for, are usually used as classification variables prior to allocation. This can easily be done for one or two variables by dividing the observed range(s) of the key classification variable(s) into sub-groups and ensuring, as far as possible, that treatments appear in all groups (as in balanced designs) or some treatments or subjects are not disadvantaged as in unbalanced or in unequal replication scenarios. Several conflicting allocation results are likely if done independently for more than 2 important variables. This paper suggests the use of a multivariate procedure to achieve an effective allocation of treatments. The method is illustrated with data on 24 animals from 2 breeds of crossbred dairy cows for allocation to 6 treatment groups (2 diet types and 3 levels of draught work). An extension to the use of the method is provided for selection of socio-economic units.