This month’s paper from Science is not the usual traditional science fare I’ve tended to blog about. I heard about this on the Science magazine podcast (yes, I subscribe to it). In it, two economists basically find a way to run a randomized clinical trial to see what microfinance does!
For those of you who don’t know what microfinance is, the idea is actually pretty simple. In many developing countries, the banking system is underdeveloped. And, even if a mature banking system were to exist, banks themselves typically do not lend small amounts of money to small businesses/families who don’t have much by the way of credit history. Enter microfinance. The idea is that you can do a lot to help people in developing countries by providing their smallest businesses, especially those run by women who are traditionally excluded from their local economies, with small (or “micro” :-)) loans. Organizations like Kiva have sprung up to pursue this sort of work, and the 2006 Nobel Peace Prize was even awarded to a man, Muhammad Yunus, because of his role in microfinance.
But, does it work at building communities and improving economies? If you’re a scientist, to answer that question conclusively, you need a controlled experiment. So, the authors of the study worked with a for-profit microfinance organization in the Philippines, First Macro Bank (FMB), to do a double-blinded randomized trial. Using a computer program, they automatically categorized a series of microcredit applicants by their creditworthiness. Obviously credit-worthy and obviously credit-unworthy applicants (combined, 26% of applicants) were taken care of quickly. For the 74% of applicants that the program considered “marginal” (not obviously one way or the other), they were randomly assigned to two groups: a control group that did not receive a microloan, and a treatment group who would receive a microloan. Following the “treatment”, the participants in the experiment were then surveyed by along on a number of economic and lifestyle metrics, and their results compared.
How was this double-blinded? Neither the applicants nor the FMB employees who interfaced with were aware that this was an experiment. The surveyors were not even aware this was an experiment or that FMB was involved.
Why focus on “marginal” applicants? A couple of reasons: first, the most likely changes to microfinance policy will impact these applicants the most, so they are the most relevant group to study. Secondly, you want to try to make apples-to-apples comparisons. Rejecting some obviously credit-worthy (or credit-unworthy) individuals may have raised red flags that some sort of algorithmic flaw or artificial experiment was happening. To really understand the impact of microfinance, you need to start on even footing in a realistic setting (esp. not comparing obviously credit-worthy individuals with so-so- credit-worthy individuals)
So, what did the researchers find? They found a lot of interesting things, actually – many of which will require us to re-think the advantages of microfinance. The data is presented in a lot of boring tables so, unlike most of my science paper posts, I’m not going to cut and paste figures, but I will summarize the statistically significant findings:
- Receiving microfinance increases amount of borrowing. The “treatment group” had, on average, 9% more loans from institutions (rather than friends/family) than the control group (excluding the microloan itself, of course)
- Microfinance does not seem to go towards aggressive hiring. The “treatment group” had, on average, 0.273 fewer paid employees than the control group. Whether or not this reflected the original size of the businesses is beyond me, but I am willing to give the researchers the benefit of the doubt for now.
- Microfinance does not seem to have a major impact on subjective measures of quality of life except elevated stress levels of male microfinance recipients. Most of the subjective quality of life measures showed no statistically significant differences except that one.
- Receiving microfinance reduces likelihood of getting non-health insurance by 7.9%
- There don’t appear to be significantly different or larger impacts of microfinance on women vs. men.
So, when’s all said and done, what does it all mean? First, it appears that instead of leading to aggressive business expansion as it is widely believed, microfinance itself actually seems to have a small, but slightly negative impact on employment at those businesses. While I don’t have a perfect explanation, combining all the observations above would suggest that the main impact of microfinance is not business expansion so much as risk management: entrepreneurs who received microloans seemed more willing to consolidate their business activities (i.e., firing “extra” workers who might have been “spare capacity”), avoid buying insurance, and reach out to other banks for more loans — very different than the story that we usually hear from the typical microfinance supporter.
The fundamental unknowns of this well-crafted study, though, are around whether or not these findings are that useful. While the researchers did an admirable job controlling for extraneous factors to reach a certain conclusion for a certain set of people in the Philippines, its not necessarily obvious that the study’s findings hold true in another country/culture. These studies were also conducted a few months after receipt of the microfinance — it is possible that the impacts on businesses and local communities need more time to manifest. Finally, the data collected from the study does almost too good of a job stripping out selection bias. Microfinance organizations today can be fairly selective, picking only the best entrepreneurs or potentially coaching/forcing the entrepreneurs to allocate their resources differently than the mostly hands-off approach that was taken here.
All in all, an interesting paper, and something worth reading and thinking about by anyone who works in/with microfinance organizations.
Paper: Karlan et al., “Microcredit in Theory and Practice: Using Randomized Clinical Scoring for Impact Evaluation.” Science 332 (Jun 2011) – doi: 10.1126/science.1200138