Suppose we want to test whether an experimental drug reduces blood pressure more than does a placebo. We are planning to administer the drug or the placebo to some subjects and record how much their blood pressures are reduced. We have 20 subjects available.
(a) We could form 10 matched pairs, where we form a pair by matching subjects, as best we can, on the basis of age and sex, and then randomly assign one subject in each pair to the drug and the other subject in the pair to the placebo. Explain why using a matched pairs design might be a good idea.
(b) Briefly explain why a matched pairs design might not be a good idea. That is, how might such a design be inferior to a completely randomized design?