Imagine you are planning to make a charitable donation and want your gift to make a real difference. You’ve done your research and found three very effective programs: one provides cash transfers to increase the incomes of very poor households; one provides treatment to correct clubfoot, a congenital condition that causes pain and mobility loss; and one provides children with medication to prevent illness and death from malaria. How do you decide?
Like GiveWell, you may aim to maximize the amount of “good” your donation does. But how can you compare giving opportunities when the good that programs do often looks very different?
Let’s go back to the three options you found. We very roughly estimate that a $10,000 donation might do one of the following:
- Double the consumption of around 15 very poor people for one year1This is based on only the direct effects of providing unconditional cash transfers that allow people to purchase more goods and services; it does not account for spillover effects or health benefits.
- Correct clubfoot for eight children
- Prevent the death of two children from malaria
Estimating the impact of your donation in this way doesn’t resolve the dilemma of where to give—it just raises other important questions. Which of those outcomes has the greatest impact? Is preventing a child’s death more or less valuable than doubling a household’s income? How much more or less valuable? In order to make a decision about where to donate, you’ll need a way to compare very different outcomes.
GiveWell faces the same set of questions. Because our funding is limited, we have to make choices about which programs to support, but those programs don’t all have the same outcomes. We need a way to compare programs to each other.
Assigning moral weights
We do this by using what we call “moral weights,” units of value that we assign to different outcomes. In our cost-effectiveness models, we use moral weights to compare outcomes and help us understand how much a particular program is likely to help people per dollar donated. We have established values for three key outcomes:
- Mortality: How valuable is preventing a person’s death?
- Morbidity: How valuable is reducing someone’s suffering?
- Consumption: How valuable is increasing someone’s ability to purchase what they need?
Over GiveWell’s history, we have done ongoing work to determine values for these various outcomes, and our approach has changed over time.
At one point, we invited all staff to come up with a set of moral weights, then combined their resulting impact estimates in our calculations. Then, in 2019, we funded IDinsight to survey approximately 2,000 people living in poverty in Ghana and Kenya. The survey asked them about their preferences regarding averting deaths at different ages and trading off averting a death with increasing income. The results suggested that, when compared to the moral weights developed by GiveWell staff, the respondents valued averting death more highly than increasing consumption.
Combining the results of that survey with the other work we had done, we assigned a value of 1 to doubling consumption for one person for one year and 100 to averting a death. This basic ratio continues to inform our moral weights. We also found that those participating in the survey valued the lives of children much more highly than the lives of adults. This same pattern showed up when we surveyed our donors. In order to account for this, we now place the greatest value of the life of a five- to nine-year old, with comparatively lower values for people who are younger or older (see here for the range of values and an explanation of how we came up with them).
When a program’s outcome is reducing illness or suffering, we base our moral weights on YLDs (“years lived with disability”), a widely used global health metric equivalent that represents the equivalent of one full year of healthy life lost due to disability or ill health. We assign a value of 2.3 to averting one YLD, then weight particular conditions based on their severity.2We calculated the moral weight of averting one YLD by dividing the value of averting a death at each particular age by the average remaining years of life at that age, then averaging the results across ages. See here for more details.
A simplified example
Let’s look at a highly simplified example to see how these values are used as we evaluate funding opportunities.
Imagine that GiveWell is evaluating a program that prevents people from dying of a disease. If the program saves the lives of 100 people and each life is hypothetically valued at 100 units, the program would generate 10,000 units of value (100 lives x 100 units each = 10,000 units).
Now imagine instead that GiveWell is evaluating a program that prevents mobility impairments for 100 people, and the benefits last for 30 years. To determine the value of that program, we’d start by assessing the severity of the impairment based on values (between 0 and 1) from the Institute for Health Metrics and Evaluation, a research institute that collects and analyzes global health data. We’d then multiply this by the number of people, the number of years during which those people benefit, and our moral weight for averting a YLD. Let’s say the program prevents “severe motor impairment,” to which the IHME assigns a value of around 0.4.3IHME defines severe motor impairment as “is unable to move around without help, and is not able to lift or hold objects, get dressed or sit upright.” The value of the program would be 0.4 x 100 people x 30 years x 2.3 per YLD, for an overall impact of 2,760 units.4Our actual calculation of a program’s value would also include a discount rate to benefits occuring in the future in order to account for intertemporal uncertainty, such as possible fundamental changes that would render the program ineffective.
Assuming these two programs have the same overall cost, the first program is substantially more cost-effective according to our current moral weights, as it produces more than three times as much impact per dollar. (Again, this is a highly simplified example; our actual analyses incorporate a wide array of other factors.5For example, to determine the value of a program that averts deaths, we need to know the mortality rate associated with the particular condition being addressed in the specific location where the program is being implemented, how much we estimate the program will reduce mortality, the percentage of people we expect will have access to the program because of our funding who wouldn’t otherwise, and much, much more. In addition, programs that avert mortality usually have other benefits, such as decreasing illness or increasing children’s long-term income, so we also calculate those benefit streams and add them to the overall value of the program. And if those benefits extend over time, we apply a discount rate to account for factors like changes in consumption and intertemporal uncertainty. Despite all those considerations, our models are only very rough approximations. While they are useful for comparing programs, we think the specific details are quite uncertain.)
We’re not satisfied
We have invested a lot of time working on these questions, but we’re not satisfied with our current approach. It isn’t fully grounded in a clear rationale, it only accounts for three primary outcomes, and we know that many people (including people on GiveWell’s staff!) value the outcomes much differently.
We’re working on it: we recently updated our moral weights page to more clearly spell out our current approach and open questions that might lead us to change our mind. In addition, we published a report describing our efforts to model the value of contraception, some of our open questions, and considerations for grantmaking.
We also think it’s important to get more input from people who are demographically similar to the people affected by the programs we fund, so we recently funded IDinsight to review recent scholarly literature on how people in low-income countries make trade-offs between health, income, subjective well-being, and contraception, and to pilot new approaches for eliciting information about these trade-offs.
We also know that more work won’t resolve what are ultimately ethical judgments. Historically, we’ve directed more funding to health than to livelihoods programs in part because our current moral weights—which consider an averted death around 100 times more valuable than doubling a person’s income—make it harder for income-focused programs to meet our cost-effectiveness threshold. But we don’t view those weights as the only reasonable standard, so we’re considering how to better incorporate moral frameworks that place greater importance on economic empowerment (for more on this effort, see our recent podcast episode on livelihoods programs).
To allow you to explore the impact of your own values, we have developed a moral weights tool where you can enter your own moral weights to see how that affects our Top Charity programs’ cost-effectiveness.
You might disagree
Translating the suffering and death of real people into units of value can seem callous. Unfortunately, we don’t have enough funding to avert all preventable deaths or address all preventable suffering. We need a way to decide where to allocate scarce resources, and we want to allocate them in a way that has the greatest impact.
Our moral weights help us in that effort, but there are no easy—or absolute—answers to the questions they help us navigate. You might assign more (or less) weight to increasing consumption. You might value outcomes, such as autonomy, that are not included in our moral weights. That’s why, in line with our core value of transparency, we publish the full details of our analysis, including our judgment calls and tough moral tradeoffs: so you can evaluate our work and decide for yourself.
Notes
| ↑1 | This is based on only the direct effects of providing unconditional cash transfers that allow people to purchase more goods and services; it does not account for spillover effects or health benefits. |
|---|---|
| ↑2 | We calculated the moral weight of averting one YLD by dividing the value of averting a death at each particular age by the average remaining years of life at that age, then averaging the results across ages. See here for more details. |
| ↑3 | IHME defines severe motor impairment as “is unable to move around without help, and is not able to lift or hold objects, get dressed or sit upright.” |
| ↑4 | Our actual calculation of a program’s value would also include a discount rate to benefits occuring in the future in order to account for intertemporal uncertainty, such as possible fundamental changes that would render the program ineffective. |
| ↑5 | For example, to determine the value of a program that averts deaths, we need to know the mortality rate associated with the particular condition being addressed in the specific location where the program is being implemented, how much we estimate the program will reduce mortality, the percentage of people we expect will have access to the program because of our funding who wouldn’t otherwise, and much, much more. In addition, programs that avert mortality usually have other benefits, such as decreasing illness or increasing children’s long-term income, so we also calculate those benefit streams and add them to the overall value of the program. And if those benefits extend over time, we apply a discount rate to account for factors like changes in consumption and intertemporal uncertainty. Despite all those considerations, our models are only very rough approximations. While they are useful for comparing programs, we think the specific details are quite uncertain. |
Comments
The “moral weights” approach and all of the effort made to come up with a way to address the difficult trade-offs are very useful for those of us who also struggle with this dilemma. Adding in animal welfare/suffering makes the choices even more difficult. I appreciate all of the thought and consideration that the GiveWell team has given to this matter.
There is one factor that I think is missing from the measurement of the value or moral weight of the death of a child, though I do not know how to quantify that factor. The loss of a child has a tremendous impact on the parent (especially the mother, but also the father and other close family members). As a physician I see parents who are still grieving many years after the death of their child. Prolonged Grief Disorder (included in DSM-5-TR) causes substantial prolonged suffering, which should be taken into account when weighing the moral value of interventions that save the lives of children.
I could be wrong, but I’d honestly guess that the majority of the moral weight for the death of a young child is attributed to the suffering of their parents. This might sound extremely callous, but the death of a newborn baby — only considering the loss of life itself — seems to be ethically equivalent to that of an abortion, or not procreating in the first place., at least from a consequentialist point of view. The net effect is the same — a life that could have been lived was not.
Hi Angelo,
We appreciate you raising this question over the years and sharing your experience as a physician. Our moral weights are largely based on individuals’ stated preferences, but we don’t model the emotional or social impact of death explicitly or separately. In theory, asking for people’s stated preferences provides the opportunity for a parent, in your example, to include their own perception of the emotional impact of losing a child in their estimates. However, it’s difficult to know exactly what people are taking into account, and there are many factors for people to consider. In the initial beneficiary preferences survey we funded in 2019, a moderate number of individuals talked about the emotional burden of losing a child when exhibiting the preference for life-saving programs (it wasn’t the most common qualitative reason people gave, but it came up repeatedly). You can read more about the report here.
As usual, GiveWell presents a well-discussed and carefully thought out roundtable on facing moral decisions. Even the everyday person with just $25:or $50 has to often choose from a seemingly endless barrage of pleas and requests for financial help, especially since so many groups make a good portion of their income from selling your info once you donate. More requests come in since you, as a donor, are now a “live” one. But what happens when you give with good intentions and then find that it is blocked from being delivered or administered. Your money is there, the goods are there, the personnel and volunteers are there, but a country’s military or a rebel group puts a stop to it all. On the moral weight scale, there could perhaps be a minus factor put in…if 10,000 points is calculated but cannot be utilized, then the actual carried weight would return to zero. Yet another harsh reality…
Great to see you keep working on this!
One reflection from thinking about this (while doing cash transfer, health, livelihoods and other work over the past decade). I wonder if you can sense check and perhaps help get to a reflective equilibrium of sorts by also comparing to choices that people are actually making (in the communities where GiveWell wants to have impact), for themselves and for others. For example, in terms of actual spending – how do people use cash transfers, and how do they allocate household budgets? And on the government side, not just VSL calculations but how are elected leaders actually allocating discretionary funding, and what are they campaigning on? And then to the extent that this differs from survey data or expectations, do we understand why?
Some of this may not fully reflect preferences due to poor information, due to corruption, etc., but I do think there are things to learn from these expressed preferences about things we may be missing in survey design, similar to how the studies on GiveDirectly’s work dug in on ROSCAs and the economic case for roofing investments after seeing the spending there. Or on the government side – after GD cash transfers led people to replace grass roofs, the local MP launched a program called “Ondoa Nyasi” to better respond to what he saw his voters wanted); we might also observe growing school feeding spending and think about how this ties to weights / preferences. If you went more down this path, one thing I’d be curious about are how to reconcile higher actual household / govt investment in education / livelihoods vs. health with some of these survey findings on health priorities. But anyway, hope you keep at it!
Hi Mark,
Thanks for sharing these thoughts! This is a great suggestion. We agree that looking at “revealed preferences”—how people and governments actually spend their money—is a really important complement to the “stated preferences” data we get from surveys, and we’re actively working on it.
Our Research team has been reviewing the literature on revealed preferences, thinking about how we could incorporate these findings into our moral weights framework, and considering how much weight to place on them.
If you have specific thoughts on potential methodologies, we would welcome your ideas—please send them to our team at research@givewell.org. Thanks again for engaging with this work!