Manuel Bueno

Moving Beyond Case Studies: Randomized Experiments

To most readers the name Esther Duflo may not ring a bell at all. However, those who have received some formal training in development economics will instantly know who she is. Esther Duflo is probably one of the most important people in development economics at barely the age of 35. Currently a professor at MIT (where she got her PhD in Economics) Duflo has seen her work published in the most important academic journals, has received a huge amount of academic honors such as the highly prestigious Elaine Bennet Prize for Research by the American Economic Association in 2003 and has been included as one of the 100 top public intellectuals by Foreign Policy. More recently The Economist named her as one of the most promising young economists.

How has she made such a splash in such a short time? In two words: randomized experiments. The basics of randomized experiment methodology are relatively straight forward. A researcher selects a group of people for a study. A random selection of this group receives a treatment (such as vaccination, better quality classes or a loan) and a control group receives either a placebo or nothing. Both groups of people are examined before and after the treatment. Since the individuals belonging to both groups are similar any differences between them after the treatment are assumed to be due to the treatment. By comparing both groups we can get a more exact idea of the impact of the treatment. This statistical technique is very similar to that used when conducting clinical trials and, if properly conducted, is one of the easiest to interpret, since it involves comparing the means from the treatment and the control group, but one of the most powerful, because it conclusively proves causality.

Duflo started applying this simple methodology to a whole array of development issues to find out exactly what worked and what not. Many of the things she found out have turned out to be extremely relevant for policy (and often economically counterintuitive). For example, mothers in the in the Indian state of Rajasthan were three times as likely to have their children vaccinated if they were rewarded with a kilogram of lentils (worth 50 cents) at the immunization camp. In 2003 alongside with Abhijit Banerjee she founded the influential Abdul Latif Jameel Poverty Action Lab which is devoted to running, promoting the use of, and disseminating the results of randomized evaluations of poverty programs – by the way, Banerjee also filmed the interesting documentary, “The name of the disease” in 2006 (which can be watched in YouTube) for those interested in health issues.

So why could randomized experiments be important for the BoP field? CK Prahalad, in his recent interview at WRI, advocated in favor of case studies as a way of generating momentum behind the idea that business can be a partner in the fight against poverty, while being financially self-sustainable. However, in addition to case studies and as this field becomes increasingly established, the BoP community needs to come up with more elaborate methodologies. BoP entrepreneurs and BoP organizations need to be perceived as legitimate vehicles in the fight against poverty by development groups, intergovernmental organizations and funding institutions.

To scale up business operations, BoP investors need to be able to conclusively show to their donors and shareholders that their investments are generating a change that would have not happened otherwise. Acumen Fund is at the forefront in the standardization of measurement tools and has pioneered many investment benchmarks that are now starting to be adopted by other BoP funds – however, if the BoP community wants to be perceived as a member of equal status in the development arena we need to go further than this. I argue that the next level involves using randomized experiments.

We at the BoP claim that businesses can be a force for good and assist in the economic development of the private sector and the communities in which these businesses are embedded. We advocate in favor of business solutions to poverty. Conducting randomized experiments is one of the best ways out there to argue that BoP businesses are more than just a collection of well-run firms and idealistic entrepreneurs, and show that BoP businesses can tackle one of the hardest problems in development: sustainability and scalability. While randomized experiments have become quite common in health or education economics, they are virtually unknown in microenterprises.

Fortunately, such studies are now starting to come out. One of the first ones (Mel, McKenzie and Woodruff, 2008) found that grants of US$100 and of US$200 in Sri Lankan small retailers and manufacturers generated average real returns of between 4.6% to 5.3% per month (or 55% to 63% per year). Although the authors did not evaluate the social returns of the businesses, they did prove that, firstly, returns can be high at low levels of capital stock (and so there was no trace of the much touted poverty trap in this case) and, secondly, that the returns could be very variable (20% of the enterprises had returns below the market interest rates). The highest returns to capital were found for poor, urban, high ability (as measured by education or a digit span recall test), male microenterprise owners. A similar outcome was found in a smaller study in Mexico (McKenzie and Woodruff, 2009).

Of course one could argue that a grant is not a loan and that the fact that people knew that they were part of an experiment may have shaped their behavior. Still, I believe that it is a promising start. Successful randomized experiments have a lot to offer to many BoP actors. To BoP investors it will legitimate their work towards other development organizations and boost the flow of incoming funds from private and public donors. To BoP firms seeking to increase the scale of their operations it will prove to socially-minded investors the benefits of their approach. To researchers, policy makers and academics it will enhance their understanding of how well and in which circumstances do BoP solutions work and may be better than other options.

The industries that may be most interesting for this methodology are those that generate the data for those receiving the treatment and for those that do not (to be able to make a comparison). For example, in the education sector a BoP enterprise may collect data on students which have joined the program and students which have not. In the health sector, there are usually detailed records about the mortality rates of those who seek medical help (and which type of help) versus those who do not. In the financial sector, there is abundant information on accepted and rejected borrowers. Agricultural BoP enterprises may also find it relatively easy to collect the data, since neighbors often use different farming products. Moreover, those organizations working in countries which collect relatively good quality statistics, such as China and India, may have less trouble in conducting such experiments.

For those interested in getting deeper into this I suggest as a starting point “Using Randomization in Development Economics Research: A Toolkit” by Duflo, Glennerster and Kremer (2006). Moreover, if anyone wants to exploring this venue, feel free to drop me a line and I will be more than happy to give a hand.

In sum, as the Business at the BoP concept matures we need to show not only that it makes a difference, but by how much. Only that way will we be able to argue that our solutions can sometimes be better than other development efforts and join the mainstream. I believe that randomized experiments could hold the key to gaining that legitimacy within the development community and really fulfilling the promise that our approach holds.