Esther Duflo reminiscenses

I first contacted Esther (and Abhijit) in 2006, when I wanted to write a New York Times column on their RCT work in India, specifically Hyderabad.  They were both extremely welcoming of my inquiries and did everything possible to give me a chance to observe their work up close.

I ended up traveling to Hyderabad, India, and spent a whole day with their RCT program in the field.  Annie Duflo, Esther’s sister, was gracious enough to travel with me around the city for an entire day, visiting the meetings where the women would show up to receive loans, and talking with the loan suppliers.  Overall I was astonished at how well-organized the work was, and how sophisticated the on-the-ground implementers were.  This was really work very carefully done.

Then, in 2013, seven years later, Banerjee and Duflo and co-authors Glennerster and Kinnan created a paper with the core results from the experiment, here is one version of the abstract:

This paper reports results from the randomized evaluation of a group lending microcredit program in Hyderabad, India. A lender worked in 52 randomly selected neighborhoods, leading to an 8.4 percentage point increase in takeup of microcredit. Small business investment and profits of pre-existing businesses increased, but consumption did not significantly increase. Durable goods expenditure increased, while “temptation goods” expenditure declined. We found no significant changes in health, education, or women’s empowerment. Two years later, after control areas had gained access to microcredit but households in treatment area had borrowed for longer and in larger amounts, very few significant differences persist.

Along with some (broadly consistent) results from Dean Karlan, this became one of the definitive papers on the effects of micro-credit.  It meant that micro-credit is OK, but not the cure for poverty.  That had a big subsequent impact on both policy and philanthropy.

You might have thought they would rest there, but no, they kept on looking at the data more deeply and over additional years, hoping to learn yet more from the experiment.  And just this last week, a new paper came out, modifying the earlier results, based on more years of data.  Here is the new abstract and paper (with Breza and Kinnan):

Can microcredit help unlock a poverty trap for some people by putting their businesses on a different trajectory? Could the small microcredit treatment effects often found for the average household mask important heterogeneity? In Hyderabad, India, we find that “gung ho entrepreneurs” (GEs), households who were already running a business before microfinance entered, show persistent benefits that increase over time. Six years later, the treated GEs own businesses that have 35% more assets and generate double the revenues as those in control neighborhoods. We find almost no effects on non-GE households. A model of technology choice in which talented entrepreneurs can access either a diminishing-returns technology, or a more productive technology with a fixed cost, generates dynamics matching the data. These results show that heterogeneity in entrepreneurial ability is important and persistent. For talented but low-wealth entrepreneurs, short-term access to credit can indeed facilitate escape from a poverty trap.

That is a pretty stunning extension of the original results, bravo to all hands involved!  Rust never sleeps, and in the hands of Banerjee and Duflo, neither does science.


Isn't this latter result just p-hacking? I didn't read the paper but searching through old data to find some positive sub-section of the population that benefited from the intervention looks pretty suspicious to me.

The results seem intuitive and replicable though. If the availability of micro-loans motivated you to start a business, you’re less likely to be successful. It’s like those guys on Shark Tank that go “uggggghhhh” when asked why they need money to do something that is not capital intensive and that they should have done already.

Yes it is. I'm glad that yours is the first comment. As noted by Alex when discussing the effect of policing on crime using the distribution of federal funds as an instrument, if you mine the same dataset several times, you'll probably come up with something significant. I think Kremer's prize is a good one. Duflo and Banergee's is less deserving.

From what I understand this wasn't a large dataset that was gathered for another purpose. It was gathered from different places, in a random controlled trial structure to see the effects of micro financing. So if there was an equal cohort of entrepreneurs who didn't get the small loans vs those who did, this isn't p hacking, which is trolling through data sets finding correlations.

Onebone) Ooga booga booga
Twobone) Hunga bunga
Threebone) Foopa dumbo

Replicable - microcredit results? Are we talking about the same literature?

Incidentally, I recall a certain blogger recently posting about the dangers of repeated analyses of fixed old datasets increasing false positive rates...

Replicable in the sense that a new experiment can be designed with the aim of testing the “gung ho” hypothesis, and that I think it would be confirmed.

The most serious complaining about this award, aside from people who think the prize simply should not exist, is that John List was not on the list. I have seen people pointing out that the wording of the award carefully emphasized the development applications of RCT rather than the field experiments aspect of it, presumably leaving the door open for List down the road.

Another criticism would be that this is too close to last year's award, especially the one for Kremer, some of whose important papers very clearly follow directly on that of one of last year's winners, Paul Romer.

"we find that “gung ho entrepreneurs” (GEs), households who were already running a business before microfinance entered, show persistent benefits that increase over time. "

Without intervention, the "market" would have rewarded most of these entrepreneurs. The intervention gives "more than market" rewards, by giving them sooner.

This is almost certainly right. "God helps those who help themselves". The market rewards success. If a method gives success faster to those who were going to be market successful, that's good.

Where is the further experiment of giving micro-credits to JUST those who have already been doing their own business?

Comments for this post are closed