I forgot to say OPISâ survey did not look into the types of pain defined by the Welfare Footprint Project.
Vasco Grilođ¸
Karma: 6,903
Thanks, Derek. What do you think about what I proposed here?
If one puts weight w on the welfare range (WR) of humans relative to that of chickens being N, and 1 - w on it being n, the expected welfare range of:
Humans relative to that of chickens is E(âWR of humansâ/ââWR of chickensâ) = w*N + (1 - w)*n.
Chickens relative to that of humans is E(âWR of chickensâ/ââWR of humansâ) = w/âN + (1 - w)/ân.
You [Carl Shulman] are arguing that N can plausibly be much larger than n. For the sake of illustration, we can say N = 389 (ratio between the 86 billion neurons of a humans and 221 M of a chicken), n = 3.01 (reciprocal of RPâs [Rethink Prioritiesâ] median welfare range of chickens relative to humans of 0.332), and w = 1â12 (since the neuron count model was one of the 12 RP considered, and all of them were weighted equally [I think RP only used 7 or 8 models for the final welfare ranges, and not neuron counts, but my point does not depend on the weight]). Having the welfare range of:
Chickens as the reference, E(âWR of humansâ/ââWR of chickensâ) = 35.2. So 1/âE(âWR of humansâ/ââWR of chickensâ) = 0.0284.
Humans as the reference (as RP did), E(âWR of chickensâ/ââWR of humansâ) = 0.305.
So, as you said, determining welfare ranges relative to humans results in animals being weighted more heavily. However, I think the difference is much smaller than the suggested above. Since N and n are quite different, I guess we should combine them using a weighted geometric mean, not the weighted mean as I did above. If so, both approaches output exactly the same result:
E(âWR of humansâ/ââWR of chickensâ) = N^w*n^(1 - w) = 4.49. So 1/âE(âWR of humansâ/ââWR of chickensâ) = (N^w*n^(1 - w))^-1 = 0.223.
E(âWR of chickensâ/ââWR of humansâ) = (1/âN)^w*(1/ân)^(1 - w) = 0.223.
The reciprocal of the expected value is not the expected value of the reciprocal, so using the mean leads to different results. However, I think we should be using the geometric mean, and the reciprocal of the geometric mean is the geometric mean of the reciprocal. So the 2 approaches (using humans or chickens as the reference) will output the same ratios regardless of N, n and w as long as we aggregate N and n with the geometric mean. If N and n are similar, it no longer makes sense to use the geometric mean, but then both approaches will output similar results anyway, so RPâs approach looks fine to me as a 1st pass. Does this make any sense?
Thanks, Ariel.
Iâd love to try surveying the general population with thought experiments to find peopleâs empirical tradeoffs of pain levels.
It looks like the Organisation for the Prevention of Intense Suffering (OPIS) is looking into this:
Suffering survey
We are currently working on a quantitative overview and visualisation of world suffering, including both humans and non-human animals. As part of this study, in September 2024 we launched a survey to assess intensity and duration of suffering in humans across a range of conditions and situations. The survey also asks questions about the nature of the suffering and any measures that were found to be effective in relieving it. The survey will remain open at least until the end of 2024.The survey is mainly multiple-choice, takes about 5-15 minutes to complete, and can be filled out anonymously. Anyone who has experienced significant suffering, physical or psychological, in the past or present, can participate, and provide information on 1-3 life conditions. We would be grateful if you would consider participating and also sharing it widely within your network!
This is the survey link: https://ââdocs.google.com/ââforms/ââd/ââe/ââ1FAIpQLSfMDXXSA-6MtPlDhhbzVv8XYIh6zvXbZcqeZJBPbHwMBIIhww/ââviewform
I have just completed it, and it took me 15 min.
Thanks for the analysis, titotal. Since you discuss some of my work, I would have appreciated it if you had shared a draft of your post before publishing it, or at least had told me when you were going to publish it.
Vascoâs figure of 1500 times as effective appears to be a result of having DALY conversion factors that are roughly ten times what RP uses, along with not discounting for less effective future campaigns.
On the other hand, I used the chicken-years per $ of broiler campaigns instead of that of both broiler and cage-free campaigns, which contributes towards underestimating cost-effectiveness. As I say in the post:
Note Saulius estimates broiler and cage-free campaigns affect 41 chicken-years per dollar, 2.73 (= 41â15) times as much as the broiler campaigns on which I am relying.
In a subsequent post, I used the chicken-years per $ of both broiler and cage-free campaigns discounted for a lower future cost-effectiveness, and concluded corporate campaigns for chicken welfare are 1.51 k times as cost-effective as GiveWellâs top charities.
I arrived at a cost-effectiveness of corporate campaigns for chicken welfare of 15.0 DALY/â$ (= 8.20*2.10*0.870), assuming:
Campaigns affect 8.20 chicken-years per $ (= 41*1/â5), multiplying:
Saulius Ĺ imÄikasâ estimate of 41 chicken-years per $.
An adjustment factor of 1â5, since OP [Open Philanthropy] thinks âthe marginal FAW [farmed animal welfare] funding opportunity is ~1/â5th as cost-effective as the average from Sauliusâ analysis [which is linked just above]â.
An improvement in chicken welfare per time of 2.10 times the intensity of the mean human experience, as I estimated for moving broilers from a conventional to a reformed scenario based on Rethink Prioritiesâ median welfare range for chickens of 0.332[6]. [Note I should have ideally used an improvement in chicken welfare per time referring to hens and broilers, instead of just broilers. I may post a cost-effectiveness analysis with separate results for broiler and cage-free corporate campaigns in the future.]
A ratio between humansâ healthy and total life expectancy at birth in 2016 of 87.0 % (= 63.1/â72.5).
In light of the above, corporate campaigns for chicken welfare are 1.51 k (= 15.0/â0.00994) times as cost-effective as TCF [GiveWellâs Top Charities Fund].
1 year of annoying pain = 0.01 to 0.02 DALYs
1 year of hurtful pain = 0.1 to 0.25 DALYs
1 year of disabling pain = 2 to 10 DALYs
1 year of excruciating pain = 60 to 150 DALYs
I think RPâs assumptions underestimate the badness of severe pain. If 1 year of excruciating pain is equivalent to e.g. 94.9 DALY (= (60*150)^0.5), 15.2 min (= 24*60/â94.9) of excruciating pain neutralise 1 day of fully healthy life, whereas I would say adding this much pain to a fully healthy life would make it clearly negative. Here is how the Welfare Footprint Project defines excruciating pain (emphasis mine):
All conditions and events associated with extreme levels of pain that are not normally tolerated even if only for a few seconds. In humans, it would mark the threshold of pain under which many people choose to take their lives rather than endure the pain. This is the case, for example, of scalding and severe burning events. Behavioral patterns associated with experiences in this category may include loud screaming, involuntary shaking, extreme muscle tension, or extreme restlessness. Another criterion is the manifestation of behaviors that individuals would strongly refrain from displaying under normal circumstances, as they threaten body integrity (e.g. running into hazardous areas or exposing oneself to sources of danger, such as predators, as a result of pain or of attempts to alleviate it). The attribution of conditions to this level must therefore be done cautiously. Concealment of pain is not possible.
The global healthy life expectancy in 2021 was 62.2 years, so maybe one can roughly say that a child taking their live due to excruciating pain would loose 50 years of fully healthy life. Under my assumptions, 0.864 s of excruciating pain neutralise 1 day of fully healthy life, so 4.38 h (= 0.864*50*365.25/â60^2) of excruciating pain neutralise 50 years of fully healthy life. However, I guess many people take their lives (if they can) after a few seconds (not hours) of excruciating pain. So, even if people should hold excruciating pain a few orders of magnitude longer to maximise their own welfare, my numbers could still make sense. 4.38 h is 5.26 k (= 4.38*60^2/â3) times as long as 3 s (a few seconds). One complication is that people may be maximising their welfare in taking their lives because excruciating pain quickly decreases their remaining healthy life expectancy, such that there is a decreased opportunity cost of taking their lives.
However, I now realise welfare ranges conditional on sentience were apparently not considered. I will ask Vicky about this.
Vicky confirmed welfare ranges conditional on sentience were not considered. So AIMâs cost-effectiveness estimates in WPs/â$ are not comparable across species, and I guess ones with lower welfare ranges conditional on sentience were overrated in AIMâs analyses (namely, weighted factor models).
I have updated the post adjusting AIMâs estimates based on Rethink Prioritiesâ median welfare ranges. The conclusion qualitatively remains:
In particular, in terms of the 5th percentile, which is the stat I considered arguably best proxying AIMâs marginal cost-effectiveness in each area, animal welfare is 48.7 times as cost-effective as global health and development.
Thanks for sharing, Ben. Here is the page where one can apply.
Thanks for the comment, David.
I canât speak for OP but I thought the whole point of its âworldview diversification bucketsâ was to discourage this sort of comparison by acknowledging the size of the error bars around these kind of comparisons, and that fundamentally prioritisation decisions between them are influenced more by different worldviews rather than the possibility of acquiring better data or making more accurate predictions around outcomes. This could be interpreted as an argument against the theme of the week and not just this post :-)
The necessity of making funding decisions means interventions in animal welfare and global health and development are compared at least implicitly. I think it is better to make them explicit for reasoning transparency, and having discussions which could eventually lead to better decisions. Saying there is too much uncertainty, and there is nothing we can do will not move things forward.
the possibility a particular intervention has a positive or negative or neutral impact on the welfare of a fish is guesswork seems very reasonable and very unfavourable to many animal related causes
What do you think about humane slaughter interventions, such as the electrical stunning interventions promoted by the Centre for Aquaculture Progress? âMost sea bream and sea bass today are killed by being immersed in an ice slurry, a process which is not considered acceptable by the World Organisation for Animal Healthâ. âElectrical stunning reliably renders fish unconscious in less than one second, reducing their sufferingâ. Rough analogy, but a human dying in an electric chair suffers less than one dying in a freezer?
Relatedly, I estimated the Shrimp Welfare Projectâs Humane Slaughter Initiative is 43.5 k times as cost-effective as GiveWellâs top charities. I would be curious about which changes to the parameters you would make to render the ratio lower than 1.
there are non-utilitarian moral arguments in favour of one group of philanthropic causes or another (prioritise helping fellow moral beings vs prioritise stopping fellow moral beings from actively causing harm) which feel a little less fuzzy but arenât any less contentious.
Why should one stop at the level of helping people in low income countries (via global health and development interventions)? Family and friends are closer to us, and helping strangers in far away countries is way more contentious than helping family and friends. Does this mean Dustin Moskovitz and Cari Tuna (the funders of Open Philanthropy) should direct most of their resources to helping their families and friends? It is their money, so they decide, but I am glad they are using the money more cost-effectively.
I guess there are sound reasons why people could conclude that AW causes funded by OP were universally more effective than GHW ones or vice versa, but those appear to come more from strong philosophical positions (meat eater problems or disagreement with the moral relevance of animals) than evidence and measurement.
One does not need to worry about the meat eater problem to think the best animal welfare interventions are way more cost-effective than the best in global health and development. Neglecting that problem, I estimated corporate campaigns for chicken welfare are 1.51 k times as cost-effective as GiveWellâs top charities, and Shrimp Welfare Projectâs Humane Slaughter Initiative is 43.5 k times as cost-effective as GiveWellâs top charities.
Thanks for the discussion, titotal and Ariel!
I have played around with Rethink Prioritiesâ (RPâs) cross-cause cost-effectiveness model (CCM), but I have not been relying on its results. The app does not provide any justification for the default parameters, so I do not trust these.
titotal, I would be curious to know which changes you would make to my cost-effectiveness estimates of corporate campaigns for chicken welfare (1.51 k times as cost-effective as GiveWellâs top charities) and Shrimp Welfare Projectâs Humane Slaughter Initiative (HSI; 43.5 k times as cost-effective as GiveWellâs top charities) to make them worse than that of GiveWellâs top charities.
They roughly agree on the GHD bar of ~20 DALYs per $1000.
The CCM says GiveWellâs bar is 0.02 DALY/â$ (as above), but I think it is around 0.01 DALY/â$. According to Open Philanthropy, âGiveWell uses moral weights for child deaths that would be consistent with assuming 51 years of foregone life in the DALY framework (though that is not how they reach the conclusion)â. GiveWellâs top charities save a life for around 5 k$, so their cost-effectiveness is around 0.01 DALY/â$ ( = 51/â(5*10^3)). Am I missing something, @Derek Shiller?
My very tentative guess is that this may be coming from Vascoâs very high weightings of excruciating and disabling-level pain, which some commenters found unintuitive, and could be driving that result. (I personally found these weightings quite intuitive after thinking about how Iâd take time tradeoffs between these types of pains, but reasonable people may disagree.)
Yes, I think this is a big part of it. From RPâs report on How Can Risk Aversion Affect Your Cause Prioritization? (published in November 2023):
1 year of annoying pain = 0.01 to 0.02 DALYs
1 year of hurtful pain = 0.1 to 0.25 DALYs
1 year of disabling pain = 2 to 10 DALYs
1 year of excruciating pain = 60 to 150 DALYs
Using the geometric mean of each of the ranges, I conclude HSI is 48.8 times as cost-effective as GiveWellâs top charities, i.e. 0.112 % (= 48.8/â(43.5*10^3)) as high as originally. I think RPâs assumptions underestimate the badness of severe pain. If 1 year of excruciating pain is equivalent to 94.9 DALY (= (60*150)^0.5), 15.2 min (= 24*60/â94.9) of excruciating pain neutralise 1 day of fully healthy life, whereas I would say adding this much pain to a fully healthy life would make it clearly negative. Here is how the Welfare Footprint Project defines excruciating pain (emphasis mine):
All conditions and events associated with extreme levels of pain that are not normally tolerated even if only for a few seconds. In humans, it would mark the threshold of pain under which many people choose to take their lives rather than endure the pain. This is the case, for example, of scalding and severe burning events. Behavioral patterns associated with experiences in this category may include loud screaming, involuntary shaking, extreme muscle tension, or extreme restlessness. Another criterion is the manifestation of behaviors that individuals would strongly refrain from displaying under normal circumstances, as they threaten body integrity (e.g. running into hazardous areas or exposing oneself to sources of danger, such as predators, as a result of pain or of attempts to alleviate it). The attribution of conditions to this level must therefore be done cautiously. Concealment of pain is not possible.
The global healthy life expectancy in 2021 was 62.2 years, so maybe one can roughly say that a child taking their live due to excruciating pain would loose 50 years of fully healthy life. Under my assumptions, 0.864 s of excruciating pain neutralise 1 day of fully healthy life, so 4.38 h (= 0.864*50*365.25/â60^2) of excruciating pain neutralise 50 years of fully healthy life. However, I guess many people take their lives (if they can) after a few seconds (not hours) of excruciating pain. So, even if people should hold excruciating pain a few orders of magnitude longer to maximise their own welfare, my numbers could still make sense. 4.38 h is 5.26 k (= 4.38*60^2/â3) times as long as 3 s (a few seconds). One complication is that people may be maximising their welfare in taking their lives because excruciating pain quickly decreases their remaining healthy life expectancy, such that there is a decreased opportunity cost of taking their lives.
I think this is a great find, and Iâm very open to updating on what I personally think the animal welfare vs GHD multiplier is, depending on how that discrepancy breaks down. I do think itâs worth noting that every one of these comparisons still found animal welfare orders of magnitude better than GHD, which is the headline result I think is most important for this debate. But your findings do illustrate that thereâs still a ton of uncertainty in these numbers.
Agreed!
Good question, Michael! Strongly upvoted. Vicky commented the cost-effectiveness estimates in WPs/â$ account for the probability of sentience. However, I now realise welfare ranges conditional on sentience were apparently not considered. I will ask Vicky about this. âCross-animal applicabilityâ was one of the goals of the WPsâ system, and I assume cost-effectiveness estimates in WPs/â$ were directly compared with each other, so I believe the welfare ranges conditional on sentience should have somehow been taken into account.
I think one should rely on RPâs welfare ranges, but I believe the best animal welfare interventions would still be more cost-effective than the best in global health and development with your multipliers. These suggest the welfare ranges should be 0.258 % (= (1.5*10*10*100*1*10)^-0.5) as large. Adjusting my estimates based on this, corporate campaigns for chicken welfare would be 3.90 (= 2.58*10^-3*1.51*10^3) times as cost-effective as GiveWellâs top charities, and Shrimp Welfare Projectâs Humane Slaughter Initiative would be 112 (= 2.58*10^-3*43.5*10^3) times as cost-effective as GiveWellâs top charities.
Thanks for the post, Nick.
Juncture 3: Dismissing Neuron count (5x â 100x multiplier)
Note you have to practically disregard the models RP used to get to this range of the multiplier. âWelfare range with neuron countsâ = âweight on neuron countsâ*âwelfare range based solely on neuron countsâ + (1 - âweight on neuron countsâ)*âwelfare range without neuron counts (RPâs)â â âweight on neuron countsâ = (1 â 1/ââmultiplierâ)/â(1 - âwelfare range based solely on neuron countsâ/ââwelfare range with neuron countsâ), where âmultiplierâ = âwelfare range without neuron countsâ/ââwelfare range with neuron countsâ. So you would need the following weights on neuron counts:
For chickens, and a multiplier of:
For shrimp, and a multiplier of:
I am far less convinced that life saving interventions are net population creating than I am that family planning decreases it. Written about 10 years ago, but still one of the better pieces on this IMO is David Roodmanâs report commissioned by GiveWell.
From the abstract of David Roodmanâs paper on The Impact of Life-Saving Interventions on Fertility (written in 2014):
In places where lifetime births/âwoman has been converging to 2 or lower, saving one childâs life should lead parents to avert a birth they would otherwise have. The impact of mortality drops on fertility will be nearly 1:1, so population growth will hardly change. In the increasingly exceptional locales where couples appear not to limit fertility much, such as Niger and Mali, the impact of saving a life on total births will be smaller, and may come about mainly through the biological channel of lactational amenorrhea. Here, mortality-drop-fertility-drop ratios of 1:0.5 and 1:0.33 appear more plausible.
So it looks like saving lives in low income countries decreases fertility, but still increases population size.
From the abstract of David Roodmanâs paper on The Impact of Life-Saving Interventions on Fertility:
In places where lifetime births/âwoman has been converging to 2 or lower, saving one childâs life should lead parents to avert a birth they would otherwise have. The impact of mortality drops on fertility will be nearly 1:1, so population growth will hardly change. In the increasingly exceptional locales where couples appear not to limit fertility much, such as Niger and Mali, the impact of saving a life on total births will be smaller, and may come about mainly through the biological channel of lactational amenorrhea. Here, mortality-drop-fertility-drop ratios of 1:0.5 and 1:0.33 appear more plausible.
So it looks like saving lives in low income countries decreases fertility, but still increases population size. Because of the decrease in fertility, it may be good to downgrade the cost-effectiveness. The above would suggest multiplying it by around 0.5 (= 1 â 0.5) to 0.7 (= 1 â 0.33).
For life-saving to reduce population, it would have to reduce total fertility by more than 1 per child saved, which is extremely implausible on its face.
Why? Each bednet costs 5 $, and Against Malararia Foundation (AMF) saves one life per 5.5 k$, so 1.1 k bednets (= 5.5*10^3/â5) are distributed per life saved. I think each bednet covers 2 people (for a few years), and I assume half are girls/âwomen, so 1.1 k girls/âwomen (= 1.1*10^3*2*0.5) are affected per life saved. As a result, population will decrease if the number of births per girl/âwomen covered decreases by 9.09*10^-4 (= 1/â(1.1*10^3)). The number of births per women in low income countries in 2022 was 4.5, so that is a decrease of 0.0202 % (= 9.09*10^-4/â4.5). Does this still seem implausible? Am I missing something?
Your authorsâ interpretation is that there is no overall effect on fertility rates: âIn this case, women simply shifted the same number of births forward, leading to more births today and less in the future.â (Indeed, if you look at their data in Figure 6, there is no evidence of any reduction in total fertility, let alone a reduction as huge as would be required for your claims.) This implies increased population long term as the saved children later go on to reproduce.
That is one hypothesis advanced by the author, but not the only interpretation of the evidence? I think you omitted crucial context around what you quoted (emphasis mine):
However, the third interpretation is a tempo vs. quantum effect: perhaps the ITN distribution induced women to have more births now, but did not change the number of overall births they intended to have. In this case, women simply shifted the same number of births forward, leading to more births today and less in the future [this is the only sentence you quoted]. Therefore, it is important to view our positive fertility results as short run, one year effects, rather than the effect on completed fertility.
My interpretation is that the author thinks the effect on total fertility is unclear. I was not clear in my past comment. However, by âlifesaving interventions may decrease longterm populationâ, I meant this is one possibility, not the only possibility. I agree lifesaving interventions may increase population too. One would need to track fertility for longer to figure out which is correct.
Indeed, if you look at their data in Figure 6, there is no evidence of any reduction in total fertility, let alone a reduction as huge as would be required for your claims.
From Figure 6 below, there is a statistically significant increase in fertility in year 0 (relative to year â1), and a statistically significant decrease in year 3. Eyeballing the area under the black line, I agree it is unclear whether total fertility increased. However, it is also possible fertility would remain lower after year 3 such that total fertility decreases. Moreover, the magnitude of the decrease in fertility in year 3 is like 3 % or 4 %, which is much larger than the minimum decrease of 0.0202 % I estimated above for population decreasing. Am I missing something? Maybe the effect size is being expressed as a fraction of the standard deviation of fertility in year â1 (instead of the fertility in year â1), but I would expect the standard deviation to be at least 10 % of the mean, such that my point would hold.
Great work, JĂŠrĂŠmy!
Thanks for the post, Henry.
Unfortunately these ranges have such wide confidence intervals that, putting aside the question of whether the methodology and ranges are even valid, it doesnât seem to get us any closer to doing the necessary cost-benefit analyses.
Large uncertainty also means a high cost-effectiveness of animal welfare research which tries to decrease the uncertainty, given the high value of information.
Nice points, Emre!
d. People seem to keep forgetting that uncertainty cuts both ways. If the moral worth of animals is too uncertain, that is also a reason against confidently dismissing them.
Uncertainty also means a higher cost-effectiveness of animal welfare research which tries to decrease the uncertainty, given the high value of information.
Thanks for taking your beliefs seriously, Remmelt. Strongly upvoted[1].
Although I think the probability of human extinction over the next 10 years is lower than 10^-6.