The Unintuitive Power Laws of Giving
Why give globally? Why give money? Why health charities? Why single-issue organizations? At first glance these all seem like arbitrary choices: what if I would rather volunteer, or donate to local charities? Why does it matter? It comes down to two distributions: cost-effectiveness and income.
DALYs per $1000
This shows the cost-effectiveness of a large number of health interventions, with taller bars in cases where we can avert more death and suffering per dollar. The shape of this chart is important: while we can do a lot of good if we pick an intervention at random or support a ‘horizontal’ effort that works on everything, we can do 300 times better by picking one in the top 10%. This is why single-issue charities make sense: you can pick one that focuses on a top intervention.
(Don’t let the small bars on the left fool you: nearly every intervention on that chart is worth doing [1], some are just far more valuable than others.)
This only considers health: what about other ways of helping people? Political advocacy, development, literacy, human rights, why not them? The big thing health has going for it is that we can measure impact, which lets us choose only the best options. In other fields where we can’t measure we could end up anywhere on the impact curve.
Let’s look at another distribution:
So some people have a lot more money that other people, we knew that, right? But have a look at the scale. Someone earning at the poverty line in the US is richer than 90% of people. This is why giving globally is so powerful: small amounts of your money can mean a huge amount to people who have so much less.
Neither of these distributions are intuitive: we don’t feel that rich, and charities all seem kind of interchangeable. But understanding them can make the difference between trying to do good and really succeeding.
(I first saw these charts in a talk by Toby Ord of Giving What We Can (GWWC). The data for the first chart, DALYs per $1000, comes from the DCP2. This was a project that, among other things, compiled cost effectiveness estimates for a very wide range of health interventions. I made the chart from the csv version of the data from here, excluding the ~60 interventions (of 171) that didn’t have estimates. The second chart is straight from GWWC’s website, and you can read the details there by clicking on footnote 4.)
[1] The median intervention there is $207/DALY, which roughly means it can give someone an extra year of healthy life for $207. Which is an incredible deal, that I think most of us would jump at. And it’s less than 5 pixels high.
I also posted this on my blog.
Note that Givewell thinks there are errors in DCP2′s cost effectiveness estimate for deworming, which they say is “the only DCP2 estimate that we have enough detail on to be able to fully understand and reconstruct.” I don’t know how reliable DCP2′s estimates are in general. Naively, I’d trust Givewell’s analyses more.
But, Givewell probably expects a power law distribution for charity cost-effectiveness, too.
As a scientist I’m going to be a total wet blanket and make a broader statement agreeing with this: it’s unclear if actual effectiveness has a power law distribution, or if it’s merely the claims of effectiveness that have that distribution. Extraordinary claims require extraordinary evidence—the surprising part of the distribution is exactly the part we should be most suspicious of.
Of course, this is why we want GiveWell doing research into it. But I don’t think they should assume this distribution is true.
“it’s unclear if actual effectiveness has a power law distribution, or if it’s merely the claims of effectiveness that have that distribution”
I talked to Toby Ord some about this today. He brought up that on the right hand of the distribution there have been some interventions (ex: smallpox eradication) that are much more effective than even the best ones available now.
Unintuitive? Are the intuitions of your expected audience really so poorly calibrated?
(I was expecting something different from the title.)
I think he meant unintuitive in the sense of “not accessible by human intuition, type 1, fast thinking” not “hard to grasp upon reflection by my intended audience”
Yes. That is the sense in which the power laws seem entirely intuitive, my model (and memory) of myself at age 8 and to the median person in the culture I grew up in. I am wondering now to what extent this intuition difference can be attributed to the degree to which the memes in question were already fully disseminated among that group. Given that I have hazy memories of conversations regarding these topics from when I was (approximately) 8 the “type 1 fast thinking’ would have a handle on it by now.
From the article:
Also, the expected audience is wider than just veteran LessWrong users.
You may then deduce that I disagree with this particular claim.
Indeed I would.
The best I can say is that the distributions are indeed unintuitive to me (at first), but I’m now well trained enough that they feel intuitive. I imagine that both you and many other LessWrongers are the same. I’d give Jeff the final word on it, but I speculate given that he cross-posted this from a blog written for a much wider audience that the charge of “unintuitive” was not meant specifically to you, but to the general non-LW-ers who would be referred to this post.
It seems a bit weird to argue over whether something is intuitive or not. Intuitive is relative to the mind doing the intuition; it’s not something “out there” that something can either be or not be. Perhaps we should take a poll of LW users?
This argumentation neglects secondary effects. Sure it is easy to quantify medicine but assume you have an intervention that doubles people’s income by some means. What are the effects of the additional income? Do they buy medicine, increase hygienic standards? I don’t know, but I suppose there are impactful second-order effects at play.
GiveDirectly is trying to quantify that. GiveWell already thinks the evidence is good enough to put them in the #2 spot but Giving What We Can disagrees.
It is worth mentioning that Wealth is particularly worse distributed than income. As if a giant kneed over that income distribution and squished it even more.
I’m skeptical of the accuracy of the income distribution chart. The Wikipedia page linked to below shows GDP per capita for every country. Many countries that we think of as “poor” are actually “middle income”. For example, Malaysia is at 16K, Colombia is at 10K, Thailand is at 9K, and China is at 8K. Obviously these mean values are somewhat misleading because of inequality, but still. I lived for many years (in grad school) at an income level of only about 2x the average income of a person living in Thailand, so I think the fact that I don’t feel rich is well-grounded in reality.
http://tinyurl.com/d5n87
Edit: there is an issue with the LW markup and Wikipedia’s URL naming convention, so I included a TinyUrl instead.
I was going to say, “double the (global) median income still sounds pretty rich.” But the median income for individuals in the US is $42,693, and I would consider double that to be well-off but not Scrooge McDuck-style rich.
I’m pretty sure that’s not median individual income, but rather median household or mean individual income.
Thanks, I think you’re right. Looks like 2005 median individual income was $28,567, so it’s unlikely to have gone up so much in seven years.
Also, in politics (human rights, etc.) you are fighting against powerful organizations, which is dangerous and can reduce the impact of your contributions. In healthcare, there usually aren’t big vested interests fighting against you.
Rationally I have to question whether giving money to charities for use in third world countries is a rational decision at all. While there are obvious benefits to, say, exterminating diseases such as polio, measles, ect. I have to question whether many of these charities are even worth investing in at all. We may claim we are looking at X many years of better life, but we have to consider:
1) Does that take into account the fact that these people are now more likely to have OTHER problems, seeing as they aren’t dying from this one?
2) Are these numbers totally made up? (The answer appears to be yes)
3) Does this take into account that some lives are considerably less valuable than others?
If I am rescuing people in Africa, their median economic output is going to be far below what I would get for rescuing but a single person in the US, and moreover, the person in Africa is going to accrue additional expenses; their economic activity is unlikely to vastly outweigh the cost of sustaining them, while the person in the US is far more likely to generate a great deal of additional economic activity beyond what is necessary to sustain them.
While it may sound cynical, if there are fewer people in Africa, then that means there are fewer people who are living in poverty; if more people there starve to death today, there will be less people starving to death there in twenty years if we save them and allow them to reproduce. Moreover, if you save one person in the US and make them self sufficient, that is a huge boost. Additionally from a personal standpoint, helping out someone locally is far more likely to bring me benefits than helping out someone in Timbuktu, and, rationally, if I’m spending money and have the option of seeing benefits myself, isn’t it more rational for me to spend the money in that way, because then I can actually appreciate the benefits of my personal generosity? Doesn’t that also encourage me to be more generous in the future, and allow others to be more generous as well? Someone who can’t even make ends meet is not going to be of the same generosity as someone who is making $60k/year.
“Does that take into account the fact that these people are now more likely to have OTHER problems, seeing as they aren’t dying from this one?”
This turns out to not be very significant.
“Are these numbers totally made up?”
The numbers in the DALY chart above come from the DCP2 project’s aggregation of cost effectiveness estimates. There are definitely errors, some of them large, but the range is there. As in, many treatments have low cost effectiveness, while others, like anti-malaria nets have strong evidence for high cost effectiveness.
“some lives are considerably less valuable than others”
I disagree on that. Averting suffering is valuable regardless of who’s suffering it is. Why are you optimizing for “economic output”?
“helping out someone locally is far more likely to bring me benefits”
Helping people, locally or remotely, is unlikely to bring you benefits. If you want to help your future self you do best saving your money or investing it. Just as you do better to purchase fuzzies and utilons separately you also do better to purchase self-benefits and utilons separately.
It depends on what you’re optimizing for—whether you value helping people, or seeing people helped (or if you value both, which one you’re focusing on with the donation in question). Being rational determines how you go about getting the things you value, not what your values are.
That said, your other arguments seem to be pointing at value-added beyond helping the one person? It does make sense to take that into account. I’m not confident of this, but I suspect that it’s not enough to make giving locally be the better option, because there is such a big difference in the amount of money required to make someone self-sufficient in the richest vs. the poorest countries.
As a percentage of income, poorer people tend to be more generous. (Unfortunately, I can’t remember where I read this, and the details of what was counted as generous (giving to friends and acquaintances, or only donations). I think the study only included people in the US.)