In partial response to Mark’s question below about anthropology, I’d like to suggest that our policy-oriented readers take a look at Robin Rogers-Dillon’s fine book, The Welfare Experiments: Politics and Policy Evaluation. Robin is an ethnographically oriented sociologist (that is, she’s kind of at the end of the discipline where it starts to morph into anthro) who was a part of the MDRC team that evaluated Florida’s 2 year time limited welfare program. In an innovation for MDRC, Robin was hired to do ethnographic work on the evaluation, which cashed out to doing a lot of hanging out in welfare offices. To make a long story short, Robin found out that the program on the ground was very different than the program on the books. For example, there was supposed to be a work program at the end of 2 years for all “compliant” clients. Well, the number of people in the work program was zero. Why? They were all administratively declared non-compliant. There was supposed to be an appeals process for the non-compliant, but it was staffed by local citizens who, it turned out, really don’t like exercising state power, so they acted primarily as a cheering squad, telling women who were cut off that “you can do it!”
To put it in technical terms, ethnographical methods (and this would certainly include anthropology) are essential, where there is any ambiguity as to what the “treatment” is in a particular policy experiment. In some cases, ethnographic methods are also valuable in helping to figure out the mechanisms that connect treatments to effects, which is highly relevant if you want to figure out how to replicate the program under other conditions. This is another case (I could list others) where well-designed policy analysis requires multiple, quite different methods in order to be serious and complete.
Another excellent ethnographic study is Katz, Kling, & Liebman, "Bullets Ain't got no Name," finding that mothers in housing projects spend tremendous amounts of time keeping their children safe from crime.
It's long been obvious to me that the formal rules of welfare programs provide pretty limited information about how programs work "on the ground." For example, another MDRC study from a while ago found that sanctioning rates (cutting welfare checks as punishment for missed appointments and such) varied greatly across counties in California's GAIN program, even though they were all ostensibly under the same formal rules. But economists continue to analyze welfare reform by comparing the rules adopted by different states. So I guess it's not obvious to everyone that the written rules don't matter that much.
I forgot to say that the authors of "Bullets Ain't Got No Name," are all economists, but it's essentially an ethnographic study.