One of the things you might want to do in designing a study at a community level is look at the variability across your communities. In our college study, for instance, we're dealing with universities and the surrounding community and we match them up in pairs.

Because we have universities that might have as many as 40,000 students and others that might be as few as 10,000 students and you wouldn't want to trust entirely with a small group of them, say 14; you wouldn't want to trust a simple random assignment to condition. So we use pairing to try to reduce the amount of variability we're gonna get at baseline between the treatment and control groups.