:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/65945249/idinsightimage.0.jpg)
If you donate to charity, you are implicitly making a lot of heavy choices about “moral weights.”
Bracketing for the moment money you spend on yourself and loved ones, $1,000 that you donate to an art museum is $1,000 that you could instead donate to medical research, or malaria prevention, or a food pantry, or your alma mater. Some of the decisions we have to make about where to give are about efficiency: How many people can this food pantry feed with $1,000, compared to another food pantry? How many lives can a malaria charity save compared to a cancer charity?
GiveWell, the charity evaluator which has directed $519 million to its preferred charities over the 2010s, has to make these decisions at a grand scale. In large part the charities it recommends do one of two things: they either, like the Malaria Consortium, focus on saving lives, or, like GiveDirectly, focus on increasing the incomes of people in poor countries. And GiveWell is not alone; other groups like the World Health Organization need to make these kinds of calls in their cost-effectiveness research as well.
GiveWell has now released a survey, conducted by the group IDInsight, of extremely poor residents in Ghana and Kenya who are likely to benefit from programs like the ones that GiveWell’s top charities conduct. The survey was meant to provide input for how to weigh saving lives versus reducing poverty in ways that reflect the values of the charity’s recipients, not just its staff.
This is the first time I can remember that a major charitable organization based in a developed country but primarily serving people in developing ones has sought input from its recipients so systematically. It has the potential to change not just GiveWell’s recommendations, but how a whole variety of international aid groups make decisions about where to put their money.
GiveWell’s challenge: weighing money against lives
Here’s the issue GiveWell faces: It has to prioritize between all the charities it evaluates, and it has to do so even though those charities do vastly different things. For instance, it has long supported anti-malaria efforts through groups like the Against Malaria Foundation, which helps distribute bednets. Bednet campaigns can save thousands of lives: GiveWell estimates that $100,000 spent on Against Malaria should save about 43 lives, and it’s directed millions upon millions of dollars to that group and other similarly effective groups over the years.
But GiveWell also has to weigh those charities against charities that aren’t trying to save lives, and aren’t best evaluated as attempts to save lives. Some of its recommended charities, like Deworm the World and the three other GiveWell top charities that implement deworming programs, are public health charities that GiveWell supports in large measure because it thinks they increase incomes in poor countries. Deworming programs prevent painful illness in young children, but it also, according to a few studies, increase long-run incomes, perhaps by improving kids’ ability to learn.
So how do you weigh that against saving lives by preventing malaria? Well, it depends on the weight you put on additional income versus saving a life. For years, GiveWell staffers have struggled with what weights to deploy in the big spreadsheet they use for ranking charities, because these are just such tough, fundamental questions. In the past, they typically had staffers put in their own moral values and then used the median response as the basis for their official recommendations.
For instance, take this spreadsheet of GiveWell staffers comparing their moral weights earlier in 2019. Founders Elie Hassenfeld and Holden Karnofsky both estimated that saving the life of a child younger than 5 is worth about 50 times as much as doubling consumption for a poor person for one year; but while Elie thought saving a child under 5 was about half as valuable as saving one older than 5, Holden thought it was about a third as valuable. Research analyst Nicole Zok weighted these lives equally, while former senior research analyst Amar Radia thought saving under-5 lives was more valuable.
These are really heart-wrenching, tough decisions to make, which inevitably come down to deep questions of personal morality. But they’re necessary if GiveWell is going to rigorously compare charities that save lives to charities that reduce poverty.
“Several GiveWell staff placed higher value on above-5 lives for a variety of reasons. These included views that older individuals are more likely to be caregivers for their family or may be making key economic contributions to the community,” Catherine Hollander, a senior research analyst at GiveWell focused on outreach, says.
GiveWell used the median estimate of these and other staff inputs, but they were operating on limited information. “Little information was available about people’s preferences in the contexts in which our top charities were operating,” Hollander says. “That seemed like information worth gathering to inform our inputs, though due to the limitations of this survey and moral weights surveys in general, we might expect staff views on how much to update based on the results to vary.”
How to estimate the monetary value of a life
Regulatory agencies in the US typically use a different method, known as the “value of a statistical life.” Popularized by Vanderbilt economist Kip Viscusi, this methodology involves trying to discern the implicit value that people in a given society place on continuing to live based on their willingness to pay for services that reduce their risk of dying.
Usually, this involves a “revealed preferences” approach. A 2018 paper by Viscusi, for example, used among other data sources Bureau of Labor Statistics Census of Fatal Occupational Injuries, to measure how much more, in practice, US workers demand to be paid to take jobs that carry a higher risk of death.
This approach is useful in considering the costs and benefits of regulation, but it has significant downsides, especially if you’re comparing across countries as GiveWell is. For one thing, revealed preferences approaches relying on actual spending decisions usually imply that the value of a statistical life is greater in rich countries than poor countries. This is for the simple reason that people in the US have a lot more money to spend to avoid death than people in, say, Kenya. That shows up in the data as Kenyans exhibiting a lower “willingness to pay” not to die, but it mostly reflects economic inequality, not Kenyans placing a lower value on their lives.
Indeed, a 2011 study by recent Nobelist Michael Kremer and others estimated an extremely, arguably implausibly low value of a statistical life in Kenya using a revealed preferences approach, which is likely a result of this problem with the method.
The revealed preferences approach also implicitly assumes a degree of rationality, and agency, in career choice that may not be present in reality. Maybe people have only the faintest idea of the fatality risks of any given job, and maybe the labor market isn’t tight enough that they actually have a choice of jobs and just take the first one that becomes available.
GiveWell’s approach
So when GiveWell supported a project by the nonprofit IDInsight to gather data to inform a new set of moral weights that could better reflect the preferences of people likely to receive their charities’ services, they opted to use surveys, not revealed preferences based on spending. IDInsight actually experimented with a revealed preferences approach before concluding the method wasn’t really working in this context.
They used two different approaches. One tried to estimate the value of a statistical life by asking about what residents would be willing to pay for a hypothetical vaccine or medicine. Here’s how IDInsight explains the question:
- Introduce respondents to the scenario, in which they are told to imagine a hypothetical disease is affecting their community. Their risk of dying from this disease is 20 in 1,000.
- Introduce the vaccine/medicine (randomized) that treats this disease, and reduces this risk from 20 in 1,000 to 15 in 1,000 or 10 in 1,000 (randomized) over the next 10 years.
- Capture respondents’ willingness to pay for this vaccine/medicine. They are told they can pay in small installments of their choosing over the 10 years of the risk reduction.
IDInsight used a variety of visual aids to make sure the example was clear, and surveyed respondents to make sure they understood before answering.
In the second approach, the firm posed more straightforward questions about which benefit recipients would rather get, like, “Program A saves the lives of 6 children aged 0-5 years AND gives $1,000 cash transfers to 5 families. Program B saves the lives of 5 children aged 0-5 years AND gives $1,000 cash transfers to [X] families. Which one would you choose?”; the survey varied the value of “X” to see at what point cash started to win out.
Both of these approaches wound up estimating higher values on saving lives relative to increasing incomes or consumption, especially saving the lives of children under 5, than GiveWell’s polls of its staff implied. The Kenyan and Ghanaian respondents placed a very high value on saving lives, particularly those of young children, and a markedly lower value on increasing consumption and income.
In response to these surveys, GiveWell announced that it’s changing its weights to reflect what the respondents said, as well as to incorporate additional arguments in favor of valuing health more highly. In August 2019, GiveWell weighted averting the death of someone under 5 at 48 times the value of doubling consumption, and averting the death of someone over 5 at 85 times the value. In November 2019, they shifted to weighting lives equally, at 100 times the value of doubling consumption.
“Moral weights seems to be a highly neglected research topic,” GiveWell writes in its post on the research. They stress that the new data is very preliminary, and a stated preferences approach has downsides, just as a revealed preferences approach does. It requires respondents to have a strong understanding of probabilities, and it’s influenced by social desirability bias: respondents might feel they can’t say they would prefer to double consumption for some number of people rather than save a life, because it feels callous.
They want much more empirical research in poor countries, more research that improves the value of statistical life as an analytical construct to make it more robust, and more philosophical work on how moral weighting ought to work.
But I give them a lot of credit. I have never heard of a major Western charity that primarily serves people in poor countries, like Heifer International or UNICEF or Doctors without Borders, conducting a study this rigorous to incorporate feedback from the people they’re supposed to be helping. It’s a technocratic approach, but a deeply democratic one too, that gives recipients a strong voice in charitable provision they didn’t have before.
Sign up for the Future Perfect newsletter. Twice a week, you’ll get a roundup of ideas and solutions for tackling our biggest challenges: improving public health, decreasing human and animal suffering, easing catastrophic risks, and — to put it simply — getting better at doing good.
