1. faiVLive Background: The motivation for the faiVLive experiment is discussing what to think about microcredit impact given all the research in recent years. If you can't make it, or if you can, here's your quick cheat sheet to the recent research.
Of course it's starts with the average impact of microcredit being very modest. A Bayesian Hierarchical model look at the data confirms those findings. But there is important heterogeneity hidden within those average effects--"gung-ho" microborrowers do see substantial gains from increased access to credit. It's also true that these are mostly studies of expanding access to formal credit, not introducing it. That's hard to measure, but we can get a cleaner view of the value of credit when it gets taken away from most everyone--and that shows significant benefits, though through a somewhat unexpected channel: casual labor wages. Changes in labor wages can matter a lot for understanding the impact of a program, even entirely masking any benefits of an intervention with evidence that it makes a substantial difference in many contexts. And it's clear that changes in labor supply are quickly passed through into labor rates--in this case, the markets seem to be working fairly well. But it's not just labor markets. When microcredit affects local markets--by increasing or decreasing the supply of tradeable goods--the benefits may be substantial but mostly captured by the people who aren't using microcredit (what economists call general equilibrium effects). Which makes it all the more important to understand local market dynamics, especially when in many cases microenterprises are operating in sectors where supply exceeds demand. That being said, microcredit is a cheap intervention relative to other options. And it's possible we could increase the returns to microcredit for more reluctant microenterprise operators by boosting their aspirations. Or perhaps by doing better targeting of lending. But is it worth targeting? Households do seem to do a pretty good job of allocating access to capital to its most productive use within the household, and the gung-ho entrepreneurs are benefiting even without the expense of targeting.
2. MicroDigitalFinance and Household Finance: I suppose all of the above would qualify here as well, but here's a bunch of different new stuff, starting with the digital side of things. There are two new papers about the effects of SMEs adopting digital payments. In Kenya, an encouragement intervention led to 78% of treated restaurants and 28% of pharmacies adopting Lipa Na m-Pesa, and consequent increases in access to credit. In Mexico, a different kind of encouragement--the government distributed massive numbers of debit cards as part of the Progresa program--led small retailers to adopt POS terminals. That led to wealthier customers shifting some of their purchasing to these smaller retailers, and increased sales and profits for the retailers, but not an increase in employees or wages paid. On a side note, it's curious that the smaller shock of debit card distribution (pushing debit card ownership to 54% of households) had a large effect on retailers but the larger shock of m-Pesa being adopted by practically everyone has not led to more Lipa Na m-Pesa adoption.
A few weeks ago I featured a puzzle in savings from two savings encouragement experiments--the encouragement worked but savings plateau at a level well below what would seem optimal. Isabelle Guerin sent me a couple of papers that I'm still reviewing that might help explain why, but this week I stumbled across another example. The US CFPB, back in the days when it was allowed to do stuff and wasn't a hollow shell of existential dread, ran an experiment using American Express Serve cards and the "Reserve" functionality. They find that encouraging savings works--people boost their savings--but that the savings plateau after the 12 week encouragement and stay at roughly the same level for 16 months. That's consistent with the results from India and Chile but not with a model of accumulating lump sums or precautionary savings. You would expect among this population that they would experience a shock in that 16 month period and draw down the savings. Participants say they reduce payday loan use, but frankly I don't believe any claims about payday behavior that isn't based on administrative data (and it doesn't make sense if balances were stable).
And finally because I want to encourage this behavior, Maria May sent me an interesting new paper on offering microcredit borrowers flexibility in repayment--customers get two "skip payment" coupons to use during the term of their 12 month loan cycle. Consistent with the much earlier work from Field et al, it yields more investment from borrowers, better outcomes and lower defaults.
3. Evidence-Based Policy: I noted last week that GiveWell, where I have served on the board since it's founding, released it's Top Charity recommendations. One of those is GiveDirectly. GiveWell, as is it's wont, wrote up some details of it's analysis of GiveDirectly, particularly about spillovers from cash transfers. That analysis was significantly informed by a forthcoming paper on general equilibrium effects and spillovers from one of GiveDirectly's programs that GiveWell was given access to even though it is not yet public. Berk Ozler took issue with that. And GiveWell responded. I have nothing whatsoever to do with GiveWell's research process or conclusions, but I was heavily involved advising GiveWell on its response to Berk's questions.
All of that is interesting, but I wanted to quickly draw attention to the Evidence-Based Policy subtext: internal and external validity. As Berk noted there are a number of papers that show negative spillovers from cash transfers in other contexts, and he makes the implicit argument that those papers are more internally valid because of public scrutiny and peer review. But are they externally valid--do their findings apply in other contexts? And more specifically, how should one weight research that has not had it's internal validity "boosted" by public scrutiny but is presumably more internally valid for being a study of the actual program being considered? GiveWell is putting a lot of weight on the non-public study because it has a large sample, is randomized and is pre-registered. Of course, one of the co-authors is a co-founder of GiveDirectly, which obviously presents some conflict-of-interest concerns. But one of the other co-authors might be called a godfather of the research transparency movement (OK, I'm doing it; I'm calling Ted Miguel a godfather of the research transparency movement).
Evidence-based policy is hard.
And that's before we factor in any of the complications of working with government and trying to incorporate community voice and self-determination. Susannah Hares reviews some lessons on "how, why and when to evaluate government-led reforms" through the lens of three impact evaluations of education policy reforms, from Delhi, Madhya Pradesh and Liberia. And since I'm speaking with Karthik Muralidharan later today, here's a throwback to a discussion he kicked off with comments at a recent RISE conference on evaluating policy reforms.
4. Methods: I suppose that last link might belong more in a methods category, let's go right there. Well, let's go back to those internal validity questions. There's no shortage of discussion on the internet of whether peer review makes a difference or not. But it's much more rare to be told that "robustness checks are a joke." Double faiV points if you guess who wrote that before clicking. Quadruple points if you can guess who it links to.
On the other hand, sometimes the rigors of peer review, robustness checks and working to have your research finding integrated into policy are just too much. It would all be much easier if your findings were suitably dramatic, surprising and large, even if that's not consistent with the data you've gathered. Here are ten simple rules for faking your research and getting it published (and not retracted).
5. Global Development : Please don't interpret that last link as having anything to do with what is in this item. Esther Duflo famously has noted that an advantage of RCTs is they have the ability to surprise. For my part, I'm frequently surprised by what expermenters manage to convince people to randomize. For instance, how about randomizing the religious content of a poverty intervention delivered by a Christian charity? That happened, in the Philippines, and here's a Freakonomics podcast about it. The results indicate that the evangelical Protestant content does increase effort and earnings. That's consistent with historical work in Germany, and by the way with the wave of work on the role of aspirations and hope.
Speaking of aspirations and hope, those two characteristics would seem to be disproportionately held by migrants. Here's Michael Clemens and Katelyn Gough on the "best ideas for making migration work."
Finally, I've had a few links on updated work on the impact of the Green Revolution, which remains surprisingly controversial, hanging around waiting for the right faiV to include them, and well, I'm going to include them today. Here's a paper exploiting time variation in the development of high yielding crop varieties and their diffusion that finds that a 10% increase in use of HYVs increases GDP per capita by 15% (within a sample of 84 countries). Alternatively, here's work that finds that HYVs delayed industrialization and urbanization in India, and thereby limited GDP growth. But here's another new paper that finds that while HYV may have kept people in rural areas, it did decrease infant mortality. So if you weren't feeling bad enough about the difficulty of evidence-based policy and evaluating policy reforms as they happen, keep in mind that 30 years later people will still be arguing over whether the reform was good or bad.
Whether you survey husbands or wives about household assets matters, and can have a substantial effect on poverty measures. Via Development Impact Blog, Source: Adnan Silverio-Murillo.
