How about those at the margin? I'd guess most policy proposals don't seem productive when you condense a population down to a scalar "model" subject and ask if he would change his behavior in light of the new policy. That doesn't mean that population-level effects won't be observed.