GiveWell’s Recommendation of GiveDirectly

Will

GiveWell recently updated their recommended charities. The principal change was that they now have a third top-recommended charity, GiveDirectly, a charity that makes unconditional cash transfers to the very poor in Kenya. GiveWell gave GiveDirectly their #2 ranking - above Schistosomiasis Control Initiative but below Against Malaria Foundation.

My initial reaction to this recommendation was surprise, and, after reading the material available on their website, I remain surprised that they have chosen to upgrade GiveDirectly to the #2 slot. This post explains why. In a follow-on post, I'll take the opportunity to make some more general comments on things that I find lacking or am concerned about in GiveWell's research and recommendations. Before I begin, I'll note that their blog and website does repeatedly mention that more information will be released soon, so hopefully the issues below will be addressed there. I'll also note that GiveWell takes direct criticism from others as both helpful and a sign of respect, which is exactly how I intend what follows. Finally, I'll caveat that the views below are my own first reactions; they aren't meant to represent the views of GWWC as a whole.


Concerns about GiveWell's Recommendation

cashGiveWell mention that cash transfers face an unusually low burden of proof. If that's a low burden of proof for doing some good, then I agree. I think that cash transfers would be a reasonable standard of comparison by which to judge other charities - it would be an indictment on the charity if they were performing worse than cash transfers.

However, there are theoretical reasons why direct cash transfers should be less effective than the best intervention in the world. There are a number of ways in which we might be able to 'leverage' our resources to achieve a surprisingly large impact, discussed below. Cash transfers do not exploit any of these.

1. Positive Externalities

Let us suppose that those in extreme poverty are perfectly rational, with perfect information. Even if so, we should expect people - anyone, but especially those whose basic needs are not met - to use money in order to benefit themselves or their household rather than whatever would have the largest total benefit, including the effect on those around them. That certain goods have positive 'externalities' is not a reason for individuals to purchase them, whereas it should be a reason for donors to purchase them. So by purchasing and providing goods with high positive externalities, the best charities should be able to outperform cash transfers.

Far from being a purely theoretical point, positive externalities actually seem to be key for the best interventions. The best health interventions increase productivity, which benefits the individual but also benefits those around them. In addition, preventing one person from catching a contagious disease prevents them from passing it on to others. Furthermore, parents may not be motivated to invest the socially optimal amount in their children's health and education. Not only does a lot of the benefit of this go to their children into the far future, but higher productivity will help other future citizens of the country. J-PAL has found that income increases of the extremely poor are often spent making life more pleasant, such as by buying nicer-tasting food, rather than more, or more nutritious, food. This is discussed in an accessible way in the book Poor Economics.

GiveWell have data on how these cash transfers are spent: they could compare the positive externalities of tin roofs with healthcare. All of the points above have also been noted by GiveWell in one of their blog posts, which makes it surprising that this issue wasn't discussed in this release.

2. Economies of scale

The extremely poor can't currently buy albendazole (a deworming drug) for $0.50. Those living in rural areas simply can't access it without using considerably more than $0.50's worth of time travelling. In contrast, thanks to economies of scale, a mass deworming program is able to provide albendazole and other drugs for ~$0.50 per treatment. One could respond that the market would provide albendazole if it there was demand for it, but it's unclear whether that's true in the situations in which deworming charities operate. This is at least worth discussing.

Like the issue of externalities, this concern holds even if we assume the extremely poor to be 'perfect' economic agents. But there are predictable ways in which they are not.

3. Lack of Information

An important way in which the extremely poor differ from idealised economic agents is lack of information. How much can we expect the extremely poor to know about the health benefits of consumption of albendazole? Or even the difference between long-lasting insecticide treated bednets and untreated bednets? For the world's poorest people, reliable health information can be too expensive, or simply impossible, to access. This creates a large industry of local practitioners with no medical training who dispense unhelpful or even dangerous advice. For those who would like to learn more, this problem is also discussed in Poor Economics.

Healthcare is often regarded as a case where markets fail because the 'information asymmetry' between the consumer and supplier is so large. For this reason among others, government and non-profit providers play a much larger role in healthcare around the world than they do in most other markets. We should expect, and have plenty of empirical evidence suggesting, that the extremely poor are extremely poorly informed about medicines and healthcare. As a result, it is not foolish to hope that experts could do additional good by promoting or distributing medicines they wouldn't otherwise choose to buy.

4. Poor Rationality

This was the issue on which GiveWell focuses in their analysis, in their discussion of whether cash transfers increase alcohol and drug consumption. I personally would expect this to be a case where the poor behave closest to the standard economic model; with so little money, spending choices can be a matter of life-or-death, so the poor will likely think extensively about their purchases given the knowledge they have.

That said, the poor, like most of us, may discount future benefits more heavily than we would like. This creates a kind of positive externality: as altruists we care about their future selves as much as their present selves, whereas they might care more about their present selves. There is some evidence, again discussed in Poor Economics, that the poor, under high stress and faced with constant temptation, struggle to save as much for the future as they would like.

 

I don't take the above theoretically driven concerns to demonstrate that GiveDirectly can't be one of the best charities out there. But they do make me initially skeptical. What surprised me is that GiveWell didn't address or even mention any other than the 'rationality' concern. There was only a discussion of whether the additional cash is used to purchase alcohol or other drugs, or whether it has unintended side-effects like causing jealousy among neighbours.

In addition to the doubts raised above, I had some other concerns with GiveWell's methodology.

1. Lack of Evidence

My overall impression of GiveDirectly is that the evidence behind it is considerably more limited than either AMF or SCI. One salient instance of this is that, as far as I can tell, GiveWell rely on self-reported data in their assessment of where received cash is spent (where they suggest that 67% is spent on home improvement), in precisely a context where we would expect self-reporting to be misleading. Another is that, though there is extensive economic analysis of cash transfers, this has focussed on "income transfers" rather than "wealth transfers" (to use the distinction that GiveDirectly make), and we might see very different effects between the two.

2. Lack of discussion of cost-effectiveness

One factor that I found very surprising is that there was no almost no discussion of how beneficial the marginal expenditure of cash transfer recipients actually is. How beneficial is it to have a metal roof rather than a bednet? What are the long-term benefits, and benefits to others, of metal roofs, as compared to bednets? This is something on which GiveWell has said more information will follow, so I won't dwell on it. But given the crucial importance of the question, I'm surprised it wasn't discussed in the basic page.

More generally, I was very surprised by the limited discussion around "what do you get for your dollar?" Out of a roughly 8000-word page, only 450 words were dedicated to this question. GiveWell claims to agree with GWWC that one should give to the charities with the best "good out per dollar in" ratio; given this, I would expect them to consider "what do you get for your dollar?" to be the crucial question.

I look forward to understanding better the thinking behind this new recommendation as GiveWell releases more information on the subject. For now, at least, I'm unconvinced.