When N=many, in this case, you get self-selection bias, people are liable to be more enthusiastic about an approach they choose themselves, etc. Maybe controlling for enthusiasm is what keeps dieters in randomized intervention studies from maintaining very well after six months or so.