Why Banerjee, Duflo, and Kremer Won the Nobel Prize Yesterday and What Their Work Means Today
That Abhijit Banerjee, Esther Duflo, and Michael Kremer would eventually win the Alfred Nobel Memorial Prize in Economic Sciences has never been the subject of much debate in the development economics community. The only surprise when the prize was awarded yesterday was how soon it came. Usually Nobel prizes are awarded to septuagenarians for a lifetime of work, or a discovery made two generations ago. The tradition in economics is that new breakthroughs need to be thoroughly vetted before they are fêted. And this vetting process typically takes half a lifetime of confirming the extent to which a clever innovation of the past has woven its influence thoroughly into the economics fabric.
In this context, awarding a prize just two decades after the authors began their foray into employing randomized field experiments to understand the effects of poverty interventions is extraordinary. And it is reflected in the age of the recipients—all in their fifties or less, with Esther Duflo being the youngest economics Nobel recipient in history at age 46. I first met Esther, along with her advisor Abhijit Banerjee, when she was a 24-year-old doctoral student at MIT, presenting one of her dissertation papers on software contracting in India at a small development conference at Yale. There were qualities to her presentation, uncompromised precision with an unruffled composure that even then seemed to portended a substantial mark on the field.
It was shortly after this that she and Banerjee began to follow Michael Kremer in experimenting with experiments in poverty work. The randomized controlled trial in general was not a new idea. The randomized assignment of subjects to treatment and control groups had been used to test the effects of new pharmaceuticals for decades. But the approach had not been implemented widely in economics at all. The first well-known experiment, on the impacts on schooling of a deworming treatment, was carried out by Michael Kremer with his doctoral student Ted Miguel, now a professor and leading development economist at UC Berkeley. What this research showed was the powerful effects of randomly administering small and very inexpensive doses of albendazole to Kenyan children vastly increased their attendance at school such that one could buy increased days in school with pennies on the dollar. Work on this particular randomized trial continues to this day, where the researchers are still finding significant impacts of deworming from this experiment on cognitive ability into adulthood.
Duflo and Banerjee began to run similar types of randomized trials across an array of development programs. A full list of their papers and the impact of these is beyond the scope of a blog post, but many of these have resulted in substantial breakthroughs in our understanding of the impacts of educational interventions, gender participation, health, agriculture, and microenterprise development. Virtually all of their papers appear in top economics journals, space reserved only for the most interesting, important, and well-executed work.
But there are many economists out there, and many excellent research papers. Why was the introduction of randomized controlled trials so transformational to the field of development economics? Before the rise of the randomized trial, increases in computational power were driving economics from primarily a theory-driven field to an empirical one. But there remained significant barriers to convincing empirical work. Quite simply, it was often hard to believe the results of many empirical economics papers. In seminars and referee reports, critics of research too often were able to accuse empirical economists of confusing causation with correlation. Because economists believed at the time that it was impractical to run economic experiments, the profession was generally stuck with purely observational data, non-experimental data in which the effect of a poverty intervention was often statistically tangled up with outcomes that would have occurred anyway, intervention or not. What that meant in practical terms was that a researcher might claim to have data showing that microfinance made entrepreneurs grow businesses like bamboo shoots, but it was possible that they were the kinds of entrepreneurs who would have grown bigger businesses anyway. And the same story could have been told with educational or health programs—those self-selecting into treatment may not be comparable to those who didn’t, the old apples and oranges comparison problem.
Some of the go-to techniques of the 1990s for evaluating poverty programs were certainly better than some of the common practices at the time, which often simply compared people before-and-after being part of a poverty program, or compared program beneficiaries with non-beneficiaries. The difference-in-differences technique, which actually combines the two ideas, is actually able to discriminate between what works and what doesn’t under a special set of assumptions, but these were often not convincingly satisfied in practice. The field was craving for bullet-proof approaches for measuring true causality.
The Banerjee-Duflo-Kremer randomization revolution provided just this kind of robust technique. But there were two problems: The first was scale. Decent sized randomized trials with the power to detect real effects are expensive to carry out. The second one was “external validity”—the biggest criticism then and today of randomized trials is that the answers they yield are simply local results that may not apply to the rest of the world. There needed to be generous sources of reliable funding and lots of replications across many countries. The solution was Banerjee and Duflo’s creation in 2003 of the Poverty Action Lab with Sendhil Mullainathan, which after substantial funding by MIT graduate Mohammed Abdul Latif Jameel became “JPAL”. With this kind of institutional development able to attract tens of millions of dollars in research funds, leading development economists and their doctoral students at MIT and Harvard were up and running, yielding some of the most credible and important results seen to date on a wide array of poverty programs carried out across the developing world.
Some of these early results, such as the Miguel and Kremer work on de-worming, led to a dramatic scaling up of the intervention and even to new global movements such as the Deworm the World Initiative. Work by Duflo and Banerjee validated the holistic poverty intervention approach of the Bangladesh Rural Advancement Committee (BRAC) across seven countries. Other of their work in India and Morocco has challenged the notion of strong microfinance impacts, indeed finding that microfinance only seems to help the top 10-15% of entrepreneurs. New results from randomized trials emerge seemingly every week, and they are changing the way billions of dollars are allocated in poverty aid.
The authors continue to build on the foundation of randomized trials, now incorporating machine-learning techniques to understand not only whether poverty interventions work on average, but among what populations they are most effective. Perhaps this is part of the excitement of seeing a group of people win the Nobel at the peak rather than the twilight of their careers. It means that there is so much yet to come.
Follow AcrossTwoWorlds.net on Twitter @BruceWydick. His new book Shrewd Samaritan is available from Thomas Nelson (HarperCollins).
- Self-Assessment Tests for Shrewd Samaritan: Faith, Economics, and the Road to Loving Our Global Neighbor
- Reflections on a Self-Experiment: 24 Hours of Homelessness in San Francisco