Randomized experiments have become a popular tool in development economics research, and have been the subject of a number of criticisms. This paper reviews the recent literature, and discusses the strengths and limitations of this approach in theory and in practice. We argue that the main virtue of randomized experiments is that, due to the close collaboration between researchers and implementers, they allow the estimation of parameters that it would not otherwise be possible to evaluate. We discuss the concerns that have been raised regarding experiments, and generally conclude that while they are real, they are often not specific to experiments. We conclude by discussing the relationship between theory and experiments. The last few years have seen a veritable explosion of randomized experiments in development economics and with it, perhaps inevitably, a rising tide of criticism. Almost all of the criticism is well-meant, recognizing the benefits of such experiments while suggesting that we not forget that there are a lot of important questions that randomized experiments cannot answer. Much of it is also not new. Indeed, most of the standard objections (and some not so standard ones)

Although institutions are believed to be key determinants of economic performance, there is limited evidence on how they can be successfully reformed. The most popular strategy to improve local institutions in developing countries is “community driven development ” (CDD). This paper estimates the impact

We present results from a field experiment aimed at strengthening school committees, and subsequently improving learning outcomes, in Indonesian public schools. School committees in treatment schools receive a grant plus a combination of three interventions: training, democratic election of school committee members, or facilitated collaboration between the school committee and the village council (called linkage). We find that measures that reinforce existing school committees, grant and training, demonstrate limited effects; while measures that foster ties between school committees and other parties, linkage and election, lead to greater engagement by education stakeholders and in turn to learning. 1

Abstract: There is an inherent tension between implementing organizations—which have specific objectives and narrow missions and mandates—and executive organizations--which provide resources to multiple implementing organizations. Ministries of Finance/Planning/Budgeting allocate across ministries and projects/programs within ministries, development organizations allocate across sectors (and countries), foundations or philanthropies allocate across programs/grantees. Implementing organizations typically try to do the best they can with the funds they have and attract more resources, while executive organizations have to decide what and who to fund. Monitoring and Evaluation (M&E) has always been an element of the accountability of implementing organizations to their funders. There has been a recent trend towards much greater rigor in evaluations to isolate causal impacts of projects and programs and more ―evidence based ‖ approaches to accountability and budget allocations. Here we extend the basic idea of rigorous impact evaluation—the use of a valid counter-factual to make judgments about causality—to emphasize that the techniques of impact evaluation can be directly useful to implementing organizations (as opposed to impact evaluation being seen by implementing organizations as only an external threat to their funding). We introduce structured experiential learning (which we add to M&E to get MeE) which allows implementing agencies to actively and rigorously search across alternative project designs using the monitoring data that provides real time performance information with direct feedback into the decision loops of project design and implementation. Our argument is that within-project variations in design can serve as their own counter-factual and this dramatically reduces the incremental cost of evaluation and increases the direct usefulness of evaluation to implementing agencies. The right combination of M, e, and E provides the right space for innovation and organizational capability building while at the same time providing accountability and an evidence base for funding agencies.

HIV testing and counseling for pregnant women has been scaled-up in sub-Saharan Africa on the grounds that it enables the delivery of medicines that prevent mother-to-child transmission of the virus and promotes behavioral changes among tested women. This paper uses longitudinal data from a high HIV prevalence region to study the take-up of HIV testing and the impact of learning HIV status on a range of health and economic behaviors. We show that health-worker absence is one of the important barriers to take-up of testing among pregnant women. Using health worker absence as an instrument for the endogenous choice of getting tested, we show that learning one’s HIV status results in significantly higher probabilities of receiving medication that can prevent the transmission of HIV from mother to child as well as other valuable health services at the time of delivery. We also examine ∗This project would not have been possible without the support of the Academic Model for Prevention and Treatment of HIV/AIDS (AMPATH) and members of the IU-Kenya partnership. We are grateful to David Cutler, Rajeev Dehejia, Duncan Thomas, and seminar participants at Harvard for valuable com-

Abstract not found

Abstract: Development organizations need a mechanism to institutionalize learning. They need a management tool that helps them plan; make decisions about what resources, or inputs, to shift; which interventions to implement, and which ones to scale; and, perhaps address their biggest challenge, to create a legitimate space for failure. The current formulation of (M)onitoring and impact (E)valuation is not conducive to learning. In this paper, we introduce (e)xperiential learning, which is the process of disaggregating and analyzing existing data to draw intermediate lessons that can then be fed back into project design, over the course of the project cycle. This is not a substitute for either ―M‖ or ―E‖, but complements the usual approach to provide development organizations with a portfolio approach to learning. Specifically, we propose a seven step dynamic MeE approach that allows one to search for what works (crawl the design space) and to identify differentials in the efficacy of the project, for real-time feedback into implementation, at key decision junctures. The idea is to use the development project itself as a learning device and not as a single ―experiment‖. This learning strategy is the project level counter-part of an emerging approach to development that might be called: ―guided incremental experimentation ‖ that emphasizes that the development process is a highly complex and contingent process that can be guided by principles, but is not reducible to simple rules or programs.

What are the causes and consequences of human development? In the twenty years since the publication of the first Human Development Report (HDR), political scientists have invested a great deal of time and effort into answering this question. So what do we know? In this paper we seek to review these labors, the fruits of which can be summarized as follows. Democracy causes, but is not caused by, economic development. While total economic growth is no higher as a result of democratic institutions, they are more conducive than non-democratic alternatives to the growth of per capita income, which is an important aspect of individual well-being. Democratic institutions are also conducive to improvements in the two other essential elements of human development, longevity and knowledge- democracy has a positive effect on indicators of education and health. Given these findings, it seems pertinent to ask why democracy has such effects. Our conclusion from the literature is that the positive impact of democratic institutions stems from their provision of accountability structures. But in providing these structures, what democracy offers is the opportunity for human development. It is no guarantee of its realization, and in the absence of factors such as information and participation this opportunity can be missed.

A model of “harassment bribes, ” paid for services one is entitled to, is developed to analyze the proposal to legalize paying these bribes while increasing fines on accepting them. We explore performance as regards corruption deterrence and public service provision. Costs of verifying reports make the scheme more effective against larger bribes and where institutions ’ quality is higher. A modified scheme, where immunity is conditional on reporting, addresses some key objections. The mechanism works better against more distortionary forms of corruption than harassment bribes, provided monetary rewards can compensate bribers for losing the object of the corrupt exchange. Results highlight strong complementarities with policies aimed at improving independence and accountability of law enforcers. (JEL D73, K42, O17) I.

Abstract Randomization has emerged as preferred empirical strategy for researchers in a variety of fields over the past years. While the advantages of RCTs in terms of identification are obvious, the statistical analysis of experimental data is not without challenges. In this paper we focus on multiple hypothesis testing as one statistical issue commonly encountered in economic research. In many cases, researchers are not only interested in the main treatment effect, but also want to investigate the degree to which the impact of a given treatment varies across specific geographic or socio-demographic groups of interest. In order to test for such heterogeneous treatment effects, researchers generally either use subsample analysis or interaction terms. While both approaches have been widely applied in the empirical literature, they are generally not valid statistically, and, as we demonstrate in this paper, lead to an almost linear increase in the likelihood of false discoveries. We show that the likelihood of finding one out of ten interaction terms statistically significant in standard OLS regressions is 42%, and that two thirds of statistically significant interaction terms using PROGRESA data can be presumed to represent false discoveries. We demonstrate that applying correction procedures developed in the statistics literature can fully address this issue, and discuss the implications of multiple testing adjustments for power calculations and experimental design. While multiple testing corrections do require large sample sizes ex-ante, the adjustments necessary to preserve power when corrections are applied appear relatively small.