Replication issues in social experiments: lessons from US labor market programs

When evaluating a pilot or demonstration program, there are risks from drawing inferences from a single test. This paper reviews the experiences of replication efforts from demonstrations using randomized controlled trials in the initial evaluation and the replications. Although replications of promising programs are primarily gathered to increase sample size, replications are also used to learn if the intervention is successful for other target groups and geographic locations, and to vary some of the intervention's features. In many cases, replications fail to achieve the same success as the original evaluation, and the paper reviews reasons that have been suggested for such failures. The paper reviews what has been learned from replications where random assignment was used in six instances: income maintenance experiments, unemployment insurance bonus experiments, the Center for Employment Training program, job clubs, job search experiments, and the Quantum Opportunity Program. The paper concludes by summarizing lessons learned from the review and areas where more research is needed.