[mlyle]

The answer is somewhere in-between:

A researcher would have a hard time getting an IRB to let them build a study at a jail where the jail treats a random half of the inmates in a different way. And judicial oversight might not allow that, either. Further, it's hard to control adequately.

We're going to be stuck with time series and case control studies of changes made haphazardly. It doesn't mean we can't get better, but it's a tougher hill to climb.