Some Observations On Design Issues in Large-Scale Social Experiments

Abstract

A number of important methodological issues have emerged from recent large-scale social experiments which have an important bearing on the ultimate usefulness of such research for policy determination This paper discusses a number of these design issues in the context of the New Jersey Income Maintenance Experiment, focusing on the importance of the shift from .conventional to optimal designs for experiments with inherently costly treatments. Risks of optimal designs lie in the possible misspecification of the response surface and the exclusion of important secondary objectives which may be politically important to policy use. The conclusion is that experimentation is a costly research method which compels optimal designs in order to identify generally small treatment effects at the usual significance levels.