Now randomly pair members of the control group with members of the intervention group in such a way as to minimize total M.I do not understand what you mean. If the pairing is random, how can it also be one that minimizes cumulative M?