Fork It Over

If you have disposable income, you have a moral obligation to give some significant fraction of it to people who have a lot less than you do. The moral imperative here is to make other people better off, not to make yourself feel warm and fuzzy. The corollary to the imperative to give, then, is an imperative to spend some reasonable effort on finding how to give effectively. Giving to the symphony doesn’t count. Sorry not sorry.

Exactly how much you should donate is arbitrary. Peter Singer has a calculator that roughly increases the fraction of your income that you donate as your income goes up. It says I should donate 5% of my income. I believe nearly everyone can and should give ~10% as a floor. My belief in that arbitrary cutoff most certainly has something to do with my Mormon upbringing, although some secular groups also advocate for a 10% donation floor. 

For the rest of this post, I’ll go through two related areas of research. First, I’ll review research on what motivates people to give. In part, this is an exercise in learning how I can keep myself motivated to keep doing what I know I should. I’m particularly interested in persuading myself to keep donating to cost-effective charities. People are very susceptible to the emotional pull of ineffective charities. I am a people, too, so today when a Walgreens cashier asked me if I would “donate 25 cents for children,” I felt like a monster for saying no. How do I avoid this irrational emotional punishment?

Second, I’ll review very high-quality evidence for two ridiculously effective charities: the Against Malaria Foundation (AMF) and Malaria Consortium (MC). The leading authority on effective giving, GiveWell, maintains a short list of top charities and periodically updates which one will do the most good with a marginal extra donation today. For some time they’ve listed the AMF as their very first choice. This year, they’ve suggested MC is their first choice, but that recommendation isn’t as prominently placed or strongly worded as the recommendation for AMF in years past. Both of these charities come with GiveWell’s extremely high recommendation.

Motivation to Give

I. Identifiable Victim Effect

Staying motivated to give money to cost-effective charities is a challenge. Motivation is emotional, whereas ‘cost-effectiveness’ may be the least emotional concept on earth. Lots of charities imply or say outright that your money is “saving” a specific person. It makes sense for charities to market like that. It’s emotionally easier to give to a specific person than to any person. In one experiment, “people contributed more to a charity when their contributions would benefit a family that had already been selected from a list than when told that the family would be selected from the same list” (quote here, and a meta-analysis of 41 experiments on this so-called identifiable victim effect (IVE)). The literature is actually quite robust, with little evidence of selection bias. This is better than I expected. I would’ve thought there’s a strong belief that IVE exists, which would discourage publication of experiments with null results, which would, in turn, make the funnel plot below look asymmetrical. The nice, symmetrical funnel shape suggests this body of literature is largely free of publication bias. The effect of IVE, about a 0.1 standard deviation increase in giving, is statistically significant, but the authors describe it as “relatively small given the enthusiasm of the field for IVE.”

Screen Shot 2019-01-28 at 11.13.45 PM

If charities can raise more money just by saying a few words to identify who the money is helping, it would seem foolish not to. The problem is that ads pointing to a specific, identifiable recipient are misleading. The lead time to create the infrastructure needed to give aid to a specific person in a developing country is measured in years, so the money a charity spends today isn’t based on the money they raise today, it’s based on the money they raised years ago. Explaining this to donors is the fully transparent thing to do, but it could undermine the instant emotional gratification that may motivate the donation to begin with.

II. The Partial Effectiveness of Effectiveness

Reading statistical evidence for a program’s effectiveness is not very emotionally motivating. One great, real-life randomized controlled trial found that “adding scientific impact information” to “a standard qualitative story about an individual beneficiary” caused large prior donors to be more likely to give and to give more, but caused small prior donors to do the opposite. This result is consistent with the hypothesis that people who give trivial donations are doing it to feel good, look good or buy their way out of guilt and are not making a good faith effort to help other people. A more optimistic take is that there are good people in the world who want to give away a lot of money and are willing to do some homework to find out how to give it away best.

The AMF solves the dual challenges of motivation and transparency in a unique way. They use a FIFO system to earmark specific donations for specific projects. They never sent me any fake postcards from the supposed exact recipient of my money, but they did tell me my money was allocated to the net distribution in the Eastern Region of the Tororo District of Uganda. They even included hundreds of pages of documentation for how that distribution went. The catch is that you have to be patient. The first money I donated wasn’t distributed for two years. Even though I know the AMF is extremely effective and this sort of transparency is exactly what I want, it’s a bit demotivating to log in to make another donation and see that your donation from 18 months ago still hasn’t been spent.   

III. Peer Pressure

How else can I stay motivated to keep donating? I have a few ideas. One is to find a sense of community with the organization I’m donating to. (Un)fortunately, the AMF has a skeleton crew of seven staff members, and they’re based in Britain, so I don’t think they’ll be hosting any social hours in the Bay Area anytime soon. A second idea is to find a sense of community with the other people donating to the same causes. GiveWell is based in San Francisco and hosts semi-annual events where they update people on their research. I went to their December event and did find it somewhat motivating. A third idea is to talk more publicly about my donations. That should be a way to find some like-minded people to talk to about how great these charities are. Plus, I’d obviously be lying if I didn’t admit that having friends and family tell me I’ve done something good feels rewarding; so this is definitely a bit of virtue signaling. (I admit I’ve worried a lot about this last part before writing this post. A Christian upbringing doesn’t teach you to look too kindly on public charity and I don’t want to come off as holier-than-thou. (But isn’t saying that I’m worried about virtue signaling itself virtue signaling? On and on we go…)) The point of all this is to create a positive feedback loop where my friends and I all peer pressure each other into making more charitable contributions. There are a number of RCT’s showing that people do, in fact, give more money when they learn about other people giving money (here, here, here).

IV. Identifiable Intervention Effect

My last idea is simply to bludgeon myself with evidence until I become motivated. I hope to be like the large donors from the RCT above who gave more money when the evidence got stronger. There’s also some evidence of an ‘identifiable intervention effect’ in which, “providing tangible details about a charity’s interventions significantly increases donations to that charity.” Let the bludgeoning commence!

Where to Give

I. Against Malaria Foundation

For the last four years, I’ve sent most of my donations to the Against Malaria Foundation. Long story short, they are insanely cost-effective at preventing death due to malaria. In partnership with governments, they carry out mass distributions of long-lasting insecticide-treated nets (LLINs). GiveWell has covered the AMF’s work extremely thoroughly with a skeptical eye and a focus on the same types of evidence I’m a fan of. They summarize their findings here. I’ll give a reader’s digest version.

This 2004 Cochrane review (GiveWell summary) found 23 randomized controlled trials studying the effect of LLIN distribution in 276,000 people. Meta-analysis of five trials with 201,000 people found that mass LLIN distributions reduced childhood mortality by 5.6 per 1000 children protected by the distribution (relative reduction of 17%). This absolute number is likely lower today because the baseline rate of malaria has gone down. The author of the Cochrane review says no more RCT’s of LLIN distributions are forthcoming because, with the evidence we already have, it would be unethical to assign anyone to the control group without an LLIN.

How do programs from RCT’s perform at a larger scale? In these studies, usage rates for the nets were in the range of 60% – 80%, so this already accounts for imperfect distribution and adherence. The GiveWell summary page for mass distributions finds that usage rates from large scale programs are in the same 60%-80% range as the RCT’s. Finally, this Nature article compiles extensive evidence on actual malaria control interventions and population level statistics on the burden of malaria. They conclude

We estimate that interventions have averted 663 (542–753 credible interval) million clinical cases since 2000. Insecticide-treated nets, the most widespread intervention, were by far the largest contributor (68% of cases averted).

I put that in a blockquote to emphasize that it is a huge f@#!king deal. There are lots of giant world problems that seem intractable, but malaria is absolutely tractable. Mass LLIN distributions have prevented 450 million cases of malaria (68% * 663 million) and the AMF is one of the single most effective organizations at conducting them.

Besides the effect on mortality, having malaria as a child causes some developmental problems that can reduce a person’s lifetime income. The GiveWell summary page on malaria discusses compelling evidence that population level malaria interventions can make entire nations wealthier. So, besides keeping a large number of people alive, malaria prevention makes the living better off.

AMF has distributed 87.5 million nets with $192.9 million in donations (see the top-right counter on their homepage), for a cost of $2.20 per net. GiveWell’s more critical review calculates a cost per distributed net of approximately $4.50. The difference comes down to accounting for goods and services provided by partner organizations. AMF’s actual role is to identify places that need nets and would not get them without AMF’s help, buy and ship the nets, find partners to distribute the nets, and finally to conduct surveys of net use before, during, and after the distribution. Every six to nine months for two and a half years after a distribution, AMF will randomly visit 1.5-5% of net recipients to survey whether they are still using nets and if those nets are still functional. This auditing ensures mass net distributions maintain their very high cost-effectiveness. Barely over 1% of AMF’s spending goes to corporate overhead and it does essentially no marketing. See their spending breakdown from GiveWell’s page.

Despite all the interest in LLIN’s from foundations and governments, the limiting factor on how many nets can be distributed continues to be cold hard cash. AMF has been spending about $30 million per year but currently expects to have less than $60 million available to spend over the next three years. This means they’ll be scaling back operations unless their rate of fundraising accelerates beyond expectations. A marginal dollar that you give to them will have a real impact on their operations.

AMF’s spending is particularly high leverage because of its relationship to other net funders. The UN provides 66% of all funding through the Global Fund and UNICEF and the US government provides another 21%. After these large, slow-moving institutions have made their grants several years out, countries with a funding shortfall come to AMF to fill in the gaps. This process means that AMF money really is resulting in extra nets being distributed, rather than just displacing government funding to other, less productive uses.

GiveWell brings all this evidence together into a final model to calculate the cost per life saved. They warn, “We believe that cost-effectiveness estimates such as these should not be taken literally, due to the significant uncertainty around them. We provide these estimates (a) for comparative purposes and (b) because working on them helps us ensure that we are thinking through as many of the relevant issues as possible.” With those caveats in mind, their final calculation is that one death is averted per $4,000 donated to AMF. This $4,000 estimate comes with high uncertainty, but it’s a number we should take seriously. If you, yes you, donate $4,000 to the AMF, about two years later there will be approximately one extra kid alive in the world who otherwise would have died before their fifth birthday. You’ll never get to speak to that person. Nobody will ever be able to tell you exactly who that person is. But that person is real and that person is alive because of your donation.

II. Malaria Consortium

Malaria Consortium (MC) is a GiveWell Top Charity that fights malaria but uses something called seasonal malaria chemoprevention instead of bed nets. In December, GiveWell said that if they had extra cash, their marginal donation would go to Malaria Consortium. I don’t know anything about MC or chemoprevention so this section will be my notes on what I can learn about them.

GiveWell summarizes the evidence for chemoprovention in this post, based largely on this Cochrane meta-analysis of seven RCT’s with 12,600 participants. SMC consists of giving monthly treatments of anti-malarial medication to children under five during the malaria season. It’s targeted in areas where at least 10% of children get malaria each season and where at least 60% of the yearly cases of malaria are concentrated in a four-month period. The effect on malaria is huge: SMC versus no malaria control strategy reduces malaria by approximately 75%. This even holds true in two trials with 6,000 participants that compared bed nets plus SMC to bed nets alone. GiveWell updates this meta-analysis by (1) removing one study that gave treatment only every other month and (2) adding one study that was published after the Cochrane review. GiveWell confirms the 75% reduction in malaria. Finally, the meta-study finds a 34% reduction in overall mortality, but this doesn’t reach statistical significance and the authors don’t convert this into an absolute risk reduction (like the ‘5.6 children saved per 1000 treated’ we have for bed nets).

SMC also has some moderate spillover effects. One trial found malaria rates fell about 25% for untreated children who lived in areas where most children were treated. However, this study treated children up to 10 years old, so the herd immunity effects were probably larger than in MC’s program that only treats children up to five. This result from just one trial is much less robust than the evidence for the direct effect of treatment.

What happens to children who receive some but not all of the four planned months of treatment? GiveWell’s summary says the treatment is effective for about 4-6 weeks and after that malaria rates go back to what they were before. There’s no ‘rebound effect.’ If you take pills that prevent you from getting malaria in January, you’re not any more likely to get malaria in March than if you hadn’t taken the pills in January. In other words, getting half the treatment should be about half as good as getting the full treatment. MC found that about 92% of children in targeted areas received at least one treatment and 55% of children received all four treatments. That’s far from perfect, but it’s good enough to suggest they’re very good at running large scale programs.

MC delivers four treatments, enough for one person for one year, at a fully burdened cost of $6.93. They pay local community health workers $5-$7 per day to go door-to-door looking for children of the appropriate age. The health workers give children the first of three doses administered over five days, instruct the parents on giving the next two doses, and ask them to mark a card to record the doses they give. Unlike AMF, which is a relatively small player in the large world of bed nets, MC is the largest distributor of SMC, treating about 7 million children yearly from 2015-2017. Going forward, they plan to use money raised on their behalf by GiveWell to treat 3.8 million children in Burkina Faso, Chad, and Nigeria.

GiveWell’s final cost-effectiveness calculation finds that one death is averted per $2,300 donated to Malaria Consortium.

So how much should I give to whom?

At first glance, SMC is roughly twice as cost-effective as the AMF. But wait! We’re trying to compare two charities that are possibly orders of magnitude more cost effective than the typical charity. Correctly identifying the very best at something is a challenging statistical problem, see a great blog post on this topic here. Think of comparing a restaurant with 2 reviews and 5 stars on Yelp to another with 1,000 reviews and 4.9 stars. Which is better? If you just rank restaurants by their average review, you’ll probably find lots of restaurants with very few reviews at the top of the list. A better ranking system has two extra ingredients: a Bayesian prior and an uncertainty interval. When we first hear about a new charity, we should assume that it is probably like other charities, but with high uncertainty. Then, as we learn more about the charity, our uncertainty will narrow. If everything we learn is positive, our mean estimate of the charity’s effectiveness will go up at the same time that our uncertainty interval is shrinking. Finally, we should rank charities not by the mean estimate of their effectiveness, but by the lower end of their uncertainty interval. This scheme won’t do a great job of ranking charities at the middle of the pack, but it will do the best job at bubbling up the very best charity that has a large amount of evidence for its effectiveness.

How does the evidence for seasonal malaria chemoprevention compare to the evidence for long-lasting insecticide-treated nets? The effect size for chemoprevention is twice the size as for bed nets (34% vs. 16% reduction in all-cause mortality), but the evidence base is much larger for bed nets (23 RCT’s and 276,000 people versus 7 RCT’s and 12,600 people). Simply having more papers and more people doesn’t automatically make the evidence better, but they appear to be high-quality trials. Having very large trials like those for bed nets should give us greater confidence that the program will maintain its effectiveness when the number of people is scaled up to millions. It’s possible that the effectiveness of chemoprevention could go down when children at lower baseline risk are treated or if adherence to taking the pills goes down when there’s less oversight. To be clear, the evidence for both charities is much much higher than exists for the vast majority of charities. I’m really splitting hairs here to find the very best. We can’t directly quantify quality of evidence, so we can’t literally calculate an error bar for each charity’s effectiveness. Roughly speaking, though, the evidence base for bed nets is so much larger than that for malaria chemoprevention, I still think AMF is likely the more cost-effective charity. In the end, the way to handle this uncertainty is by diversifying. For this year, I’ll give 2/3rds of my donations to the Against Malaria Foundation and 1/3rd to Malaria Consortium.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s