Just in reply to your last point David, I'll just say that that's one way of approaching the general question of "we need to know what large numbers of will respond to some circumstance" and is the "scientific" approach of understanding things. This does potentially run up against the issue that theres too many different factors for complicated human beings. An alternative would be to focus on the question of what engineering and science is needed to be able to reliably infer the behaviour of very large scale populations of people from practically doable experiments (in terms of population size, duration, etc) involving people. In such an approach there's less need for understanding (even though understanding would be nice).