One person might be a little low one day, another person might be high, a lot of people might-- you might happen to have captured them on an average day, but overall it averages out, so you get a representative sample.
That meant those in the control group were more likely than average to be regular users, and therefore were not representative of the population at large.