Animal welfare work has the potential to be much more cost-effective than work on global poverty. While it depends greatly on how much you value a human compared to a nonhuman animal, the suffering in factory farms appears quite severe and the scale of factory farmed animals ( ~9-11B in the US, many more globally[1]) is greater than the total world population, over 10x the number of extremely poor people, and over 40x the number of people affected by malaria.
Based on this, some have suggested that the only reason to think that animal welfare doesn’t dominate global poverty is speciesism, or the belief that nonhuman animals do not have significant moral worth.
However, another reason to think that global poverty work could be more effective than animal welfare work is based on strength of evidence—we have enough evidence to know the very best global poverty intervention, but we don’t have enough evidence to know the very best animal welfare intervention.
In this article I want to take a look at what this might mean in practice—when you have strength of evidence for global poverty but not for animal welfare interventions, you likely aren’t comparing the best animal welfare intervention to the best global poverty intervention. Instead you are likely comparing the mean animal welfare intervention to the best poverty intervention.
Furthermore, this could entail that global poverty is better now[2], since there are reasons to think the mean animal welfare intervention could be quite worse than the best global poverty intervention.
Keep in mind, however, that I think it could entail this conclusion, not that it actually does. I use “X could be true” in the sense of it is possible that X or it is reasonable for some people to think X based on what we currently know. I do not use it in the sense that X is more likely than not or I believe X and you should too.
Also, even if global poverty could be better right now in the abstract, there are still many additional considerations I don’t write about here, such as thinking about marginal funding, thinking about counterfactuals, thinking about long-term flow through effects, thinking about the value of research or meta-work, etc.
My own analysis of the raw data provided shows the minimum estimates distributed with a mean of $804.78/DALY, a median of $313.50/DALY, a min of $1/DALY, a max of $5588/DALY, and a standard deviation of $1250.12/DALY. The maximum estimates are distributed with a mean of $3557.66/DALY, a median of $929/DALY, a min of $5/DALY, a max of $26813/DALY, and a standard deviation of $6738/DALY.
While there are good reasons to not take the DCP2 estimates too literally, we’re lucky there’s a large wealth of research on global health interventions which allow us to make reasonable attempts at ranking different interventions in order of their cost-effectiveness.
If we did not have this research and had to sample an intervention at random, we would end up with the mean intervention with a potential cost-effectiveness of ~$805.78-$3557.66/DALY. Using the DCP2, we can select an intervention with a potential cost-effectiveness of ~$1-5/DALY, a potential gain of over 700x!
We’re likely smart enough to exclude interventions that are likely to be quite inferior from a scale perspective, like farm sanctuaries, we still likely face a large range of potential cost-effectiveness. While we don’t know yet if the shape is logarithmically distributed like it appears to be for global health interventions[3], it seems to me that a lot of the interventions that “smart money” would pick include the possibility of no impact (being actually worthless) and net negative impact[4] (actively causing more harm per each dollar spent), even before considering their possible far-future effects.
This suggests that while the best intervention in animal rights could exceed that of global health by ~250x[5], the mean intervention could be much worse than the best intervention from global poverty. Since we don’t have enough evidence yet to pinpoint the best animal welfare intervention, we’re in the same situation as before where we are staring at an unlabeled graph, forced to pick the mean intervention.
Extending to the Far Future
At this point there’s still an open question about how to extend this to the far future. This is quite hard for reasons that Michael Dickens points out in “Are GiveWell Top Charities Too Speculative?”—while we might have a pretty good idea of what the near-term effects of GiveWell top charities are (namely, less malaria, less parasite infections, and more wealth) and we might have some idea of the medium-term effects (more economic growth, essentially no net population growth), we have no idea of the long-term effects and this could dramatically change the overall cost-effectiveness.
This undermines the “strength of evidence” approach in favor of global poverty, but there still are many plausible views one could make that suggest global poverty comes out ahead. For example, one could reasonably think that…
It’s not clear to me which of these views, if any, are correct, and I hope to explore them a lot more, including the many other view that I did not write down here. However, it is clear to me that one could reasonably think that global poverty is more effective than animal advocacy, even while agreeing that nonhuman animals have significant moral value, based on the principle of comparing the best intervention to the mean intervention.
Edit: This post originally incorrectly assumed that randomly selecting from all possible interventions would yield the median cost-effectiveness, not the mean cost-effectiveness.
Endnotes
[1]: I’ve heard numbers around 60B (for one example, from ACE, though I’ve also heard it elsewhere), but I’ve never been able to track down an authoritative citation for this (nor have I tried particularly hard). However, I don’t think the precise number is that important for this analysis.
[2]: I think this analogy can be similarly extended to most other causes where there isn’t much evidence yet to pick among a large range of potential interventions, many of which are of zero impact or net negative.
[3]: I’m pretty curious what the overall shape of interventions plotted against cost-effectiveness would look like. I think in the case of nonhuman animal advocacy there could be reasons to think that the shape could look pretty weird if there is a large possibility of net-negative effects[4] or if many messages had roughly the same outcome (for example, maybe online ads and pamphlets are roughly equally persuasive).
[5]: For example, if the true conversion rate of online ads happens to be 3%, this suggests ~144-80388 days of animal suffering averted per dollar (using the “Simple Calculator”, fixing conversions / pamphlet at 0.03). If we assume humans are worth 1-300x more than nonhumans, this crudely suggests an estimate of 0.001-220 DALY / $. Flipping that to $/DALY would be $0.004 - $1000 / DALY.
Global poverty could be more cost-effective than animal advocacy (even for non-speciesists)
Animal welfare work has the potential to be much more cost-effective than work on global poverty. While it depends greatly on how much you value a human compared to a nonhuman animal, the suffering in factory farms appears quite severe and the scale of factory farmed animals ( ~9-11B in the US, many more globally[1]) is greater than the total world population, over 10x the number of extremely poor people, and over 40x the number of people affected by malaria.
Based on this, some have suggested that the only reason to think that animal welfare doesn’t dominate global poverty is speciesism, or the belief that nonhuman animals do not have significant moral worth.
However, another reason to think that global poverty work could be more effective than animal welfare work is based on strength of evidence—we have enough evidence to know the very best global poverty intervention, but we don’t have enough evidence to know the very best animal welfare intervention.
In this article I want to take a look at what this might mean in practice—when you have strength of evidence for global poverty but not for animal welfare interventions, you likely aren’t comparing the best animal welfare intervention to the best global poverty intervention. Instead you are likely comparing the mean animal welfare intervention to the best poverty intervention.
Furthermore, this could entail that global poverty is better now[2], since there are reasons to think the mean animal welfare intervention could be quite worse than the best global poverty intervention.
Keep in mind, however, that I think it could entail this conclusion, not that it actually does. I use “X could be true” in the sense of it is possible that X or it is reasonable for some people to think X based on what we currently know. I do not use it in the sense that X is more likely than not or I believe X and you should too.
Also, even if global poverty could be better right now in the abstract, there are still many additional considerations I don’t write about here, such as thinking about marginal funding, thinking about counterfactuals, thinking about long-term flow through effects, thinking about the value of research or meta-work, etc.
The Range of Global Poverty Interventions
In “On Priors”, Michael Dickens graphed the list of cost-effectiveness estimates from the DCP2 and found an exponential curve in terms of $/DALY (blue is the minimum estimate, red is the maximum estimate):
My own analysis of the raw data provided shows the minimum estimates distributed with a mean of $804.78/DALY, a median of $313.50/DALY, a min of $1/DALY, a max of $5588/DALY, and a standard deviation of $1250.12/DALY. The maximum estimates are distributed with a mean of $3557.66/DALY, a median of $929/DALY, a min of $5/DALY, a max of $26813/DALY, and a standard deviation of $6738/DALY.
While there are good reasons to not take the DCP2 estimates too literally, we’re lucky there’s a large wealth of research on global health interventions which allow us to make reasonable attempts at ranking different interventions in order of their cost-effectiveness.
If we did not have this research and had to sample an intervention at random, we would end up with the mean intervention with a potential cost-effectiveness of ~$805.78-$3557.66/DALY. Using the DCP2, we can select an intervention with a potential cost-effectiveness of ~$1-5/DALY, a potential gain of over 700x!
Comparing to Animal Welfare Work
Comparatively, there is very little research on how to best improve animal welfare, and what research that does exist has historically lacked control groups, been statistically underpowered, and suffered from many other problems (see Section 4 of “Methodology for Studying the Influence of Facebook Ads on Meat Reduction” for a good review).
The largest scale RCT on animal advocacy to date was only powered enough to rule out a rate of 4.2% or higher at 80% confidence , though we aren’t sure how much lower the true conversion rate is or if there are other interventions with a bigger success rate. If we were to come up with some sort of DALY vs. intervention graph for animal rights, what would it look like?
We’re likely smart enough to exclude interventions that are likely to be quite inferior from a scale perspective, like farm sanctuaries, we still likely face a large range of potential cost-effectiveness. While we don’t know yet if the shape is logarithmically distributed like it appears to be for global health interventions[3], it seems to me that a lot of the interventions that “smart money” would pick include the possibility of no impact (being actually worthless) and net negative impact[4] (actively causing more harm per each dollar spent), even before considering their possible far-future effects.
This suggests that while the best intervention in animal rights could exceed that of global health by ~250x[5], the mean intervention could be much worse than the best intervention from global poverty. Since we don’t have enough evidence yet to pinpoint the best animal welfare intervention, we’re in the same situation as before where we are staring at an unlabeled graph, forced to pick the mean intervention.
Extending to the Far Future
At this point there’s still an open question about how to extend this to the far future. This is quite hard for reasons that Michael Dickens points out in “Are GiveWell Top Charities Too Speculative?”—while we might have a pretty good idea of what the near-term effects of GiveWell top charities are (namely, less malaria, less parasite infections, and more wealth) and we might have some idea of the medium-term effects (more economic growth, essentially no net population growth), we have no idea of the long-term effects and this could dramatically change the overall cost-effectiveness.
This undermines the “strength of evidence” approach in favor of global poverty, but there still are many plausible views one could make that suggest global poverty comes out ahead. For example, one could reasonably think that…
Economic growth is likely a net good and animal welfare work is undermined by not having much of an effect on that growth .
The effects of animal advocacy on long-term values shifting are not as large as believed or are particularly fragile and unlikely to propagate across generations .
The flow-through effects are unlikely to be larger than the direct effects, which, in global poverty, are more clearly good.
It’s not clear to me which of these views, if any, are correct, and I hope to explore them a lot more, including the many other view that I did not write down here. However, it is clear to me that one could reasonably think that global poverty is more effective than animal advocacy, even while agreeing that nonhuman animals have significant moral value, based on the principle of comparing the best intervention to the mean intervention.
Edit: This post originally incorrectly assumed that randomly selecting from all possible interventions would yield the median cost-effectiveness, not the mean cost-effectiveness.
Endnotes
[1]: I’ve heard numbers around 60B (for one example, from ACE, though I’ve also heard it elsewhere), but I’ve never been able to track down an authoritative citation for this (nor have I tried particularly hard). However, I don’t think the precise number is that important for this analysis.
[2]: I think this analogy can be similarly extended to most other causes where there isn’t much evidence yet to pick among a large range of potential interventions, many of which are of zero impact or net negative.
[3]: I’m pretty curious what the overall shape of interventions plotted against cost-effectiveness would look like. I think in the case of nonhuman animal advocacy there could be reasons to think that the shape could look pretty weird if there is a large possibility of net-negative effects[4] or if many messages had roughly the same outcome (for example, maybe online ads and pamphlets are roughly equally persuasive).
[4]: This would be possible if, for example, it were true that caged-free campaigns result in a lower living standard for hens (though see the response from Bollard and ensuing discussion ) or if Direct Action Everywhere’s confrontational activism approach actually drove people away from animal rights. Both of these seem plausible enough to me to introduce negative values into my confidence interval for their cost-effectiveness estimates, even if I don’t think they are more likely than not.
[5]: For example, if the true conversion rate of online ads happens to be 3%, this suggests ~144-80388 days of animal suffering averted per dollar (using the “Simple Calculator”, fixing conversions / pamphlet at 0.03). If we assume humans are worth 1-300x more than nonhumans, this crudely suggests an estimate of 0.001-220 DALY / $. Flipping that to $/DALY would be $0.004 - $1000 / DALY.