Originally posted on the effective altruism forum, and there’s some good discussion in the comments there.
(Epistemic status: Low confidence in end result. I’m not an expert, and I am likely to be overlooking something important. The methodology is informative and there are good links. I’ve tried to point out sources of numerical uncertainty.)
How cost effective is it to help people by preventing global warming? How cost-effective is it, on the margin, to remove CO2 from the atmosphere or prevent carbon from entering the atmosphere? Is it possible to get an answer denominated in cost per life saved?
Edit: Answer: It is possible, but this essay is too oversimplified to get to that answer. Based on discussion in the comments, I think this model is making an inappropriate linearity assumption that hides a steeper gradient on the margin. You might find this exercise informative, but I believe this question’s framing distorts the answer by about an order of magnitude.
Let’s look at this equation:
(Cost Per Life Saved) = (Cost Per Ton of CO2) * (Tons of CO2 Per 1°C Warming) * (Expected Degrees Warming) / (Lives Lost Due to Climate Change)
There are a lot of unknowns and many of these terms will change due to technological or social changes, but let’s try to estimate them.
Edit: TL;DR: If we guess at these numbers, it costs about \$50,000 \$750,000 per life saved if you want to save lives by eliminating carbon from the atmosphere. However, I believe this is an overestimate by a factor of 10. See here for a different analysis which yields a significantly lower cost.
Part 1: Cost Per Ton of CO2
How much money do you have to spend to make there be one less ton of CO2 in the atmosphere?
Edit: Most of these numbers are misleadingly large, and cheaper interventions appear to be available. See note at end.
This turns out to be a relatively easy question to answer because there are people actively working on producing technology to improve this number. Here’s an article about direct carbon capture, where they say the current cost is about \$600/ton, but they’re hoping to get it down to \$100/ton. The Department of Energy has the goal of reducing this to \$30/ton.
That’s the marginal cost, and if they scaled it to address all the carbon put into the atmosphere annually, there’s reason to think that the cost might be significantly lower (because of economies of scale) or higher (because they’d run out of manufacturing capacity and land).
But that’s the cost for sucking carbon out of the atmosphere once it’s already there. What about things we can do to prevent carbon from entering the atmosphere in the first place?
Climate change economists have produced a graph called an abatement cost curve, which shows a comparison of the cost of different interventions denoted in dollars per ton of carbon.
Many of these interventions have negative cost, meaning that they reduce carbon and also cost less in the long run. Switching incandescent bulbs to LEDs is a good example of that; they last a lot longer and therefore they’re cheaper. Looking at the chart, I see there are a lot of interventions in the €15-50 range.
Table 2 in this paper contains more such cost estimates, which are generally a bit more pessimistic; it notes many interventions are in the \$30-300 range. For example, “Livestock Management Policies” are estimated to cost \$71 per ton of carbon, while the Cash for Clunkers program in the US had a cost of about \$270-420 per ton of carbon, which is worse.
Here’s an article about Biden’s climate policy, which includes a particularly cost effective climate change intervention in the form of a tax credit, which is estimated to cost \$33 to \$50 per ton of carbon.
In general, the cheapest interventions will not scale to fix the entire climate problem. Some industries like steel transport are very expensive to decarbonize.
So at the end of this section, we’ve got to put a price on a ton of carbon. I’m going to go with \$30, which is representative of some of the most effective interventions we have available today. Most interventions will be more expensive than this, but some will be cheaper.
EDIT: Commenters point out that some lower cost interventions are available in the sub ten dollar range, and potentially in the sub one dollar range. I’ve updated this essay to include that. I’m going to use \$2, although some cheaper interventions may be or may become available. In particular, see the Founders Pledge on the Clean Air Task Force, page 95.
(Side note: Tons vs Tonnes. Some of you might remember that the metric tonne at 1,000kg is different from the English ton at 2,000lb, or 907kg, and you might have noticed that I was glossing over that. However, they’re pretty close and the difference is much less than the margin of error on the conclusion, so I’m treating them as the same. Sorry!)
TL;DR:\$2/ton (Edit: was \$30)
Part 2: Tons of CO2 Per 1°C Warming
How much CO2 do we need to put into the atmosphere to warm the planet by one degree celsius?
It turns out this answer is a little bit complicated because greenhouse gasses don’t warm the planet directly, instead they help the sun warm the planet. But increased carbon does increase the steady state temperature the earth will reach. Someone on Quora ran a simulation where you can see that. Their answer was 5 trillion tons.
Some environmentalists have computed a carbon “budget” to hit various warming targets. This source suggests about ~500 gigatonnes CO2 to get 1.5°C, or ~1,500 gigatonnes CO2 to get 2°C of global warming. That works out to about 2 trillion tons per degree, or 2Tt/1°C.
Other posters on that Quora question roughly agree with that estimate, from various lines of argument, so that’s the number we’ll use.
TL;DR: 2Tt/1°C
Part 3: Expected Degrees Warming
How much do we expect the earth to warm up by? This is a complicated question that I’m not qualified to address, so I’m going to present this infographic from NASA and then interpret it badly.
Looks like about 2 or 3°C degrees of warming.
TL;DR: Let’s say 2.5°C.
Part 4: Lives Lost Due to Climate Change
If we get 2 or 3°C of global warming, that will be bad, for all the reasons that climate change is bad. But how bad, specifically? Can we put a number on how bad it will be? In particular, how many people do we expect to die because of global warming?
Edit: This section obscures the difference between overall mortality due to climate change and marginal mortality due to climate change. Models suggest that harms due to climate change are nonlinear and that marginal harms will increase at higher temperatures.
This turns out to be a hard question to answer, but there are some good estimates. There are also some bad estimates, which I’m presenting also. Here are 4 methodologies, ordered from most optimistic (but still bad) to most pessimistic. Coincidentally, the most optimistic methods are the highest quality.
Detailed Multi-Factor Analysis from the WHO
The World Health Organization put out an estimate that between 2030 and 2050, 250,000 additional people will die each year because of climate change. They created this infographic:
The full report is worth reading if you’re interested in this question. They did their research and accounted for many different potential vectors of climate change. I was originally going to say that nobody has a good answer to this question, but this research is detailed and convincing.
If we project that number over a full century, it adds up to 25,000,000 people losing their lives due to climate change.
Economic Modeling
This excellent article from R. Bressler attempts to answer the question “how many tons of carbon in the atmosphere kills one person”, and their answer is “4,434 metric tons of carbon dioxide in 2020—equivalent to the lifetime emissions of 3.5 average Americans—causes one excess death globally in expectation between 2020-2100”.
If I understand the paper correctly, they use a modification of the DICE model to get an estimate that 83 million people will die between 2020 and 2100. I think they incorporated the effect of economic losses and harms due to non-death losses into this figure, which you can see in equations 7 and 8 in their paper. They’re actually summing harms out until the year 2510, but they’re using temporal discounting so they don’t weight it very heavily. I’m not sure why they chose such a long time horizon.
Quote “After surveying these 100 studies, we found that the consensus was that climate change is likely to increase future mortality rates through a number of channels including the direct effects of ambient heat16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56, interactions between higher temperatures and surface ozone formation24,43,46,52,55,56,83,84,85,86,87,88,89,90,91,92, changes in disease patterns16,26,43,45,46,49,50,52,53,54,55,93,94,95, flooding16,26,43,50,52,53,54,55,56,93,94,95, and the effect on food supply16,26,43,45,52,54,55,56,87,95,96,97,98.”
One thing I like about this paper is that they try to balance the economic benefits of industry and economic growth against the harms of climate change and conclude that 2° of warming is “optimal”.
I’m going to extend their estimate to a full century at 100 million dead.
All the Wars and Famines
This model is pure guesswork, but it’s based on how I hear people talk about their concerns about climate change.
Let’s assume that climate change will cause global unrest which leads to terrible wars, and also that it will disrupt food production and cause mass starvation.
How many people do we expect to die in these terrible wars and famines? In the 20th century, about 100,000,000 people died from famine, and a similar number died in all the wars. Let’s assume that the next hundred years will be similar, and that 100% of it can be blamed directly on climate change. This is probably an overestimate, because there might be other causes of war[citation needed].
That works out to 200 million dead over the century, or 2 million people per year on average.
Worst Case Scenarios
This article thinks a billion people will die because of climate change. It was published in the journal of Environmental Psychology and spends a lot of the article talking about how to use its results in political messaging. Here’s a quote about their methodology: “A more rigorous multivariate analysis […] will not be attempted here. Instead, I will present a big-picture, top-down estimate.” Their numbers for death estimates are summarized in their section titled “Worst-Case Estimate”, and include a variety of potential harms.
I did not find their methodology convincing, but I am including it because it was widely cited in the press. 1 billion dead over a century.
Vague, Spread Out Harms
You may have the intuition that climate change may not kill that many people, but it will reduce the quality of life of many people across the world in a way that’s worse than the deaths it causes. With the exception of Bressler, the analyses above don’t go into that.
I find this hard to estimate, and I wasn’t able to find a good study on this. Please post in the comments if you’re able to find a rigorous analysis of this!
For the sake of estimation, let’s assume that climate change affects everyone’s lives for the worse. We need a number in the end, so let’s say that everyone becomes 5% worse off. Some people won’t be affected that much, some people will be affected worse, but 5% on average. For reference, this study (page 44-45) indicates that having a tension headache or shingles causes a 5% decrease in well-being. Following the logic of QALYs, we can say that everyone being 5% worse off is sort of equivalent to 5% of people dying, or about 500 million deaths. This might be on top of the deaths estimated by some other method.
Temporal Discounting
(Side note about time: I think temporal discounting is appropriate here, not because we care less about people in the future, but because we have less ability to predict how we affect them. Using a too-conservative but convenient 1% annual discount rate and the formula for the sum of an infinite geometric series, 1/(1-0.99), we find that the infinite sum is equal to 100 years, helpfully letting us use one century as a stand-in for all the rest of time.)
Summary
Putting these together, we have a range of death estimates from 25-1,000 million deaths over the course of a century. I think numbers toward the lower end of that range have the best support from good research, but that numbers toward the higher of that range match up better with environmentalist intuitions.
TL;DR: I’m going to err on the low side because it’s supported by better research and use an estimate of 200 million deaths.
Math: Putting it All Together
Math time! Remember our formula:
(Cost Per Life Saved) = (Cost Per Ton of CO2) * (Tons of CO2 Per 1°C Warming) * (Expected Degrees Warming) / (Lives Lost Due to Climate Change)
Let’s put in numbers:
(\$50,000 / life) = (\$2/T) * (2 Tt/°C) * (2.5°C/AGW) / (200 million lives / AGW)
(Note: AGW stands for anthropogenic global warming, and it’s a dummy unit indicating that 1 global warming will have occurred. It reflects my failure to denominate harms in deaths/°C.)
The answer we get is \$50,000 per life saved through spending to reduce global warming by removing carbon from the atmosphere.
Edit: See conclusion.
Obviously, there’s a lot of uncertainty here. The environmental science around how much warming will occur and how much CO2 will cause that warming is pretty good, and it has less than a degree of magnitude in error bars. The estimate of how many lives will be lost due to global warming is worse, and I could easily see it be off by an order of magnitude in either direction, especially when you account for non-death harms.
I’ve tried to include slightly conservative estimates throughout, so I think this number represents an optimistic, low end estimate for the cost.
Note 1: Ranges and Uncertainty
Edit: Outdated.
We saw uncertainty in each step of the model. Here’s the ranges of plausible values we found:
Cost Per Ton of CO2: \$10-\$50. This is only considering a range for the very best interventions. Most climate interventions will be more expensive than this.
Tons of CO2 Per 1°C Warming: 1.5-2.5 Tt.
Degrees of Warming: 1.5-5°C. Seems like a lot of uncertainty here.
Deaths Caused by Global Warming: 25-1,000 million.
From these we can get a low-end estimate and a high-end estimate using the formula above.
This is the best case scenario (for cost effectiveness, not for the future), where we assume that climate change will be really bad really quickly, and it’s really cheap to fix:
(\$22,500 / life) = (\$10/T) * (1.5 Tt/°C) * (1.5°C/AGW) / (1000 million lives / AGW)
This is the worst case scenario (again, for cost effectiveness), where we assume that climate change won’t be that bad and also it’s expensive to deal with:
(\$25,000,000 / life) = (\$50/T) * (2.5 Tt/°C) * (5°C/AGW) / (25 million lives / AGW)
Note 2: Climate Change is Expensive!
Edit: Climate change is still expensive, but this analysis overstates it. Cheaper interventions are available than I initially thought. However, the cheapest interventions wouldn’t scale to fix the whole problem.
One takeaway I’ve gotten from this exercise is that climate change is a really expensive problem. Reversing climate change would cost about 120 trillion dollars, or about 6 times the GDP of the USA, and slightly more than global GDP. If the US federal government spent its entire budget every year for 10 years on addressing climate change, it might be expected to address 1°C of warming, out of the 2 or 3°C that we’re expecting to experience. I can think of other ways in which we might want to spend that.
Note 3: Worst Case Scenario
Edit: Outdated
What if climate change is an existential risk that will lead to the death of all humans?
If we assume that climate change will immediately cause the death of all 8 billion humans on earth unless we pony up the money to reverse 2 degrees of warming with our best present technology, we get:
(\$15 thousand / life) = (\$30/T) * (2 Tt/°C) * (2°C/AGW) / (all 8 billion lives / AGW)
This provides an upper bound for how cost-effective climate change interventions can be.
Note 4: Comparisons to Other Cause Areas
GiveWell estimates that its top charities in global health need about \$5,000 to save one life or life equivalent, which includes quality of life impacts from illness besides death, such as injury or loss of ability to work, primarily through preventative medicine in the developing world. The cost I’ve computed above is significantly worse than that.
Edit: See commentary in conclusion, climate change may be cheaper to influence than this line of analysis suggests. Some climate change charities may be considerably cheaper than GiveWell’s top charities.
Climate change interventions might become competitive with global health charities if two things become true:
- We update our models toward increased cost-effectiveness for climate change. This might mean that climate change starts to look more harmful, or we develop much better technology to deal with it.
- GiveWell’s top charities run out of room for more funding, and the marginal available direct intervention gets much worse.
If carbon removal technology improves significantly or the impacts of climate change start to look worse, or if global health loses room for additional funding, then climate change may enter the ballpark of being comparable to GiveWell’s top charities.
If you’re optimistic about technology and pessimistic about the harms from climate change, it starts to look comparable to GiveDirectly, which is about ten times less cost effective (per life saved) than top malaria charities.
Note 5: Non Human Harms
Climate change affects animal welfare and nature in a way that malaria or global poverty don’t. A full accounting of the cost of climate change needs to account for animal suffering, but that’s outside the scope of this post. In addition, climate change is likely to reduce biodiversity, which might be seen as intrinsically valuable.
Conclusion
I did not find that climate change interventions are cost effective compared to GiveWell’s top charities, at about \$50,000 per life saved. My main source of uncertainty was in estimating the harms expected to be caused by climate change, and more research there is needed. Because climate change is so expensive to mitigate, a lower bound of about \$15,000 per life can be established in the case where climate change is treated as an X-risk.
Edit: See DannyBressler’s comment below (same Bressler cited in “Economic Modeling”). His analysis finds a considerably lower number, and his analysis is more sophisticated, so I recommend believing it over this essay’s.
Surprisingly to me, this cost is smaller than I expected. I went into this essay expecting to find that climate change would be extremely ineffective to influence, and while I still believe it’s much worse than GiveWell’s top charities, the gap is smaller than I expected.
Edit: This exercise has shifted me from believing “it is not cost effective to do anything about climate” to believing “maybe I should donate money to climate change interventions.”
I’d also like to point out that there are likely to be climate interventions that are much more effective than the \$2/ton estimate that I use. In particular, some interventions like building solar panels are likely to cost negative dollars, and it’s sort of a coordination problem to make that happen. Also, it’s likely that technical research is likely to have the leverage to drive down those costs and be much more cost effective. Estimating the cost effectiveness of research is outside the scope of this essay.
I’d like to see more analysis of this question. I think that non-expert EAs could improve on my estimate by doing a more detailed reading of the literature.