Recent academic debates pit two alternative approaches to policy evaluation against one another. The first is the "experimental" approach, based on the random assignment of accepted program applicants to a recipient, or treatment group and a non-recipient, or control group. The second is the "non-experimental," or "econometric," approach that uses a variety of microdata sources, statistical methods, and behavioral models to compare the outcomes of participants in social programs with those of nonparticipants. The central question addressed in this paper is whether or not randomized social experiments aid in securing answers to basic questions about the evaluation of social programs. The paper concludes that while experiments can eliminate the potential for selection bias to affect mean-impact estimates, the existing literature overstates many of the other arguments in their favor. There is a sizable divergence between the theoretical capabilities of evaluations based on random assignment and the practical results of such evaluations.
The paper is available to authorized users at JSTOR http://www.jstor.org
Bibliography: Journal of Economic Perspectives 9(2): 85-110
You should be able to view web pages in your web browser (Internet Explorer, Netscape, etc.)
