To estimate costs and benefits and thus provide a rational basis for decision, social policy research must forecast the responses of target populations to particular social policies and programs. Costs are mostly a function of unit costs multiplied by participation rates among eligible individuals. Benefits are measured by the distribution of net social and economic impacts. Before evaluated social experiments were available, none of these crucial predictors of policy impact could be estimated with any precision or reliability - not by one shot case studies, not by uncontrolled demonstrations, not by process evaluations, and certainly not by empirically ungrounded computer simulations. The available proven theories of socio-economic change could not, and still cannot accurately predict the behavior of large groups over years in response to new social policies and programs. But now, after ten years of increasing effective development, social experiments can do significantly better.