The Fixing People Edition
1. Global Development: In my early days of blogging global development and philanthropy stuff, the Millennium Villages Project--and specifically the controversy over claims of impact and whether to measure impact at all--were a really big deal. In a true blast from the past, the impact evaluation of a Ghanaian MVP has finally been published and found little to no impact. Little to no impact on core welfare indicators and little to no impact on spillovers or "cost-saving synergies." That impact evaluation only happened because Michael Clemens and Gabriel Demombynes went to the mat to convince DfID not to fund expansion unless it included independent evaluation. Here's a thread from Demombynes on that, and one from Clemens. It's worth noting that the MVP not only actively resisted impact evaluation but threatened Clemens and Demombynes with a lawsuit to stop their efforts.
I've been thinking about this a lot in relation to criticism of the RCT movement around the recently awarded Nobel prizes. This paper isn't an RCT--it's a Diff-in-Diff with matched villages and propensity score matching! The point is the distance we have traveled in terms of demanding credible evidence on development interventions in a very short period of time is under-appreciated. It was less than a decade ago that literally the highest profile development intervention in the world was insisting that there was no need for an impact evaluation, a control group, etc. and there was actual controversy over that position. And I do think that the randomistas are largely responsible for that change where the debate is about the relative credibility and cost-benefit of different approaches to measuring impact, and external validity of findings, not on whether to engage in credible impact evaluation.
There are other controversies from the global development past that are resurfacing, if only in Justin Sandefur's Twitter timeline. Justin--I presume because he's going to be teaching a development class at Georgetown this spring--has been asking some interesting questions and getting some interesting responses. Like, given the credibility revolution, and follow-on work, how should we think about Paul Collier's The Bottom Billion? Or how separable are Peter Singer's support for altruism, e.g. The Life You Can Save, and his support for murdering disabled babies. I guess I tipped my hand on how I feel about the latter. I can't exactly be objective here as the father of a 13 year old who, in a Singerian world, could have qualified for "elimination." Justin links to an amazing essay by Harriet McBryde Johnson that I had forgotten, but am very glad I read again--you should read it too regardless of whether you have or not. The question is one I'd been able to ignore for a long time--cognitive dissonance is powerful--and I'm grateful to have been forced to think about it again.
2. Digital Financial Services: Usually when we talk about digital financial services it's about delivering a specific financial service via digital channels. But here's a paper on digital delivery of guilt about the use of financial services. Specifically, it's an intervention where people are shown a Nollywood movie whose plot is driven by "bad" choices in relation to borrowing and saving. They find that watching the movie does induce people to open savings accounts but not to use them to, y'know, save. That's consistent with a lot of the research on savings (some of which was highlighted by the eMFP team last week). Clearly there are lots of nudges that can get people to open accounts and even to save in them, but those nudges rarely lead to meaningful ongoing use or significant savings balances.
There are exceptions of course, and here's a post from the A16Z FinTech blog that highlights a few of them. There's a common theme: savings encouragement works when it removes the consumer from the equation, or uses their bad decision making for good. Kinda dark, huh? Of note here is the idea of "self-driving money"--customizing the services and products to the needs and often irrational behavior of particular customers. It's a great concept, but there is a key question missing: where are the financial services firms that are going to automatically help a customer into a product that is less profitable but better for the customer? And in case you didn't know, financial services firms are already customizing products based on consumer biases: by sending credit card offers that are more likely to be profitable for the issuer.
On a more traditional digital financial services footing, here's a discussion of digital remittances and the lack of progress toward remittances that stay digital. I continue to find it remarkable that keeping transactions digital--e.g. not having users cash out--is assumed, without any explanation, to be obviously good for users. It's a particular case where the DFS community seems to consistently be ignoring the signals that customers are giving them. The explanations for why people don't stay in digital never seems to consider the most obvious answer: there's no benefit to the user. When there is actually benefit to customers of staying digital I have no doubt that they will do so.
3. Our Algorithmic Overlords: I've featured work by Sendhil Mullainathan and co-authors on algorithmic bias several times over the last few months. Here's a summary of that work in a highly consumable NYTimes piece, "Biased Algorithms are Easier to Fix than Biased People." Here's a thread of responses beginning with Cathy O'Neil (of MathBabe and Weapons of Math Destruction fame). Of particular note is Felix Salmon's response that access to the algorithms and underlying data is distributed highly unequally with preference to celebrated economists. I'm sure that helps, but I would guess it doesn't help all that much--regulation is the only real solution to this problem, but at least there is a solution. It's much more likely than figuring out how to de-bias people. But there is another issue that I worry about much more than whether winning a John Bates Clark medal gets you preferential access: it's that the handful of people like Sendhil and Susan Athey are among the relatively few people in the world capable of truly understanding both what the algorithms are doing and how they can be delivering biased results.
But maybe I shouldn't worry quite so much. Although I have to consider that some secret algorithm determined that the optimal way to attract my attention was to write a post about Jim Scott's Seeing Like a State, algorithms and oppressive regimes. Here the case is that reliance on algorithms is unlikely to yield the results that repressive regimes hope to achieve.
Before you relax too much, remember that the technological infrastructure that is required for algorithmic overlords doesn't necessarily just serve the interests of state actors (which is part of the reason they may not yield the desired results). The infrastructure can be repurposed for things, like, oh, spying on and talking to your children by hacking Ring cameras and then distributing the code to do so widely.
4. SMEs: The way I can tell it's been a long time since I last wrote the faiV is how thoroughly convinced I was that I hard already written about the new working paper on general equilibrium effects of GiveDirectly cash transfers in Kenya. You likely saw some coverage of the paper--The Economist, Vox--and you might be wondering why it's here under the SME heading.
That's because, far and away, I find the most interesting part of the paper to be the way the positive spillovers flow through the enterprises in the villages. Now, granted there are a lot of microenterprises and not SMEs but the basic principle applies. The cash transfers turn into increased demand for the many enterprises; that doesn't turn into inflation, apparently, because there is so much slack in these businesses. That's an incredibly important thing to understand when thinking about possible interventions to increase productivity or employment in these firms. For instance, it's a possible explanation for the muted effects of efforts to improve the business practices of small firms. If the binding constraint on their revenue isn't their practices but limited demand, no matter how much you improve those practices you're not going to get much action. Bottom line: if you do any work on SMEs you have to read this paper.
But the GiveDirectly GE paper was only the second coolest paper that I read since my last faiV. This paper by Odyssia Ng, a job market candidate from Stanford, may be the coolest paper of the year. Granted it's truly about microenterprises, but you'd have to work very hard to convince me that it doesn't apply more broadly. Ng gets to the bottom of the puzzle about why woman-owned businesses are less profitable than male-owned businesses. She sets up vegetable stalls in Jaipur, India, and keeps everything constant--hours open, product availability, negotiating ability, buyer preference--and finds that when you create this level playing field, men and women sellers earn the same. She concludes that access to capital is the driver of the revenue gap.
5. Philanthropy and Social Investment: It's the giving season, and for the first time since I can remember I didn't field any calls about Giving Tuesday--for a while I've been the token critic of Giving Tuesday for arguing it's really unclear what effect it's having, and what the goal is. That was a bit disappointing because my perspective has changed some in light of data that individual giving is falling in the US, the first time that seems to have happened when the economy is growing. I speculated some time ago that the real theory of change for Giving Tuesday was to shore up giving culture in the US which even prior to last year has shown some signs of weakening.
There's another aspect of giving season that I'm connected to: GiveWell's annual recommended charities list. That connection is I'm the chair of the board of GiveWell (and in case you were wondering about my comments about cognitive dissonance and Peter Singer, well there it is--if you click on that link you'll get an ad for a free copy of The Life You Can Save). Of course, there are critiques of GiveWell's approach. Here's a thread from Jeremy Konyndyk laying out his objections. They are quite similar to critiques of the RCT movement: GiveWell's approach necessarily leads to a focus on small, easy to measure interventions. But there's also a critique that GiveWell's emphasis on cost per life saved leads away from programs that work in the most dangerous and unstable areas. I'm strongly biased here, but neither of these points bother me much. On the first it's simply a question of whether you believe the marginal dollar of philanthropy is better spent on interventions we have high confidence in helping people or in more speculative efforts which could yield bigger gains but which we know much less about how to help. It's similar to the argument that fixing biased algorithms is easier than fixing biased people. On the second, I confess to being a bit confused. My understanding is that there are more of the extreme poor in middle income countries than there are in poor countries now and so a bias toward helping people in more stable places seems like it would yield better overall outcomes. But I'm open to more argument on that.
Finally, here's a fascinating Op-Ed from some former Gates Foundation execs on how foundations can and should change their practices in light of the "fervor of wealth taxes." That includes giving away much more than they do now. Hear, hear!
Fixing algorithms is easier than fixing people. Mostly.
