One problem is these studies often show what happens when you ask somebody to do something--as opposed to when they actually do it. This is why there are metabolic chamber studies, you can be more certain of what people are actually up to.

Compliance can be big--and one thing that I think can increase compliance is knowing that something is likely to be effective. Two equally effective interventions, but you believe one will work, and the other won't--suddenly what were otherwise equal interventions, what would give equivalent results in say a mouse or a tube-feeding study, might give very different results.