Scaling Up and Evaluation

This paper discusses the key role that impact evaluations should play in scaling up. Credible impact evaluations are needed to ensure that the most effective programs are scaled up at the national or international levels. Scaling up is possible only if a case can be made that programs that have been successful on small scale would work in other contexts. Therefore, the very objective of scaling up implies that it is possible to learn from past experience. Because programs that have been shown to be successful can be replicated in other countries, while unsuccessful programs can be abandoned, impact evaluations are international public goods: The international agencies should thus have a key role in promoting and financing them. In doing this, they would achieve three important objectives: Improve the rates of returns on the programs they support; improve the rates of returns on the programs other policy makers support, by providing evidence for the basis on which programs can be selected; build long term support for international aid and development, by making it possible to credibly signal what programs work and what programs do not work. The paper makes the case that the best way to estimate the impact of a broad class of development programs is through a randomized impact evaluation. First, it discusses the methodology of randomized evaluation through several concrete examples, mostly drawn from India. It then reviews the shortcoming of the current evaluation practices, before addressing the most frequent criticisms against randomized evaluation. Finally, it makes some suggestions about the role that international organizations could play in coordinating the accumulation of international knowledge on development projects: defining priorities for evaluation, fund and conduct randomized evaluation through an independent unit, work with partners, and create a data base of evaluation results. 1 PRELIMINARY DRAFT, FOR CIRCULATION AT THE ABCDE CONFERENCI thank Ted Miguel, and, in particular, two reviewers for extremely detailed and useful useful comments. I also thank Abhijit Banerjee, Francois Bourguignon, Anne Case, Angus Deaton and Michael Kremer for numerous helpful discussions on the issues discussed in this paper.