Until the past few years, our nation's approach to designing federal programs for preschool-age children lacked coherence and paid little attention to what had worked (and not worked) in the past. In this article, the authors propose that credible information useful for designing effective programs will require the ongoing, systematic development and evaluation of alternative approaches for the improvement of large-scale early childhood programs. The research should place greater reliance on experiments in which existing groups of individuals, such as intact classes or preschool agencies, are randomly assigned to implement competing early education programs or program components. Randomizing groups, rather than individual children, changes the research question from “What works?†to “What works better?†yielding more useful information than is currently available about which preschool approaches ought to be strongly embedded in our nation's social policy.
