The views expressed here are my own, not those of the people who provided feedback on the draft.

Summary

• Ambitious Impact (AIM) implicitly assumed all species have welfare ranges conditional on sentience equal to that of humans until 2024. Consequently, AIM had been overestimating the welfare range of species with lower welfare ranges conditional on sentience, and therefore the cost-effectiveness of helping animals of such species. For example, I believe helping fish was favoured relative to helping chickens, as fish have a lower welfare range conditional on sentience than chickens.
• Using Rethink Priorities’ (RP’s) median welfare ranges instead of AIM’s past assumption described above, the 3 fish welfare interventions AIM has recommended have a significantly lower cost-effectiveness than that of cage-free campaigns at the time of recommendation.
• I conclude influencing European Union fish welfare policy through strategic work in Greece advocating for electrical stunning before slaughter is 12.3 % as cost-effective as cage-free campaigns. So I wonder whether AIM should have recommended it in 2023, and incubated it in 2024 via the Centre for Aquaculture Progress (CAP). AIM could have relied on RP’s median welfare ranges.
• Nevertheless, I am actually glad AIM incubated CAP, as I consider they greatly underestimated how much their humane slaughter intervention decreases the time in pain weighted by intensity (not welfare range) per fish helped. AIM:
  • Estimated the welfare gain per fish-year conditional on sentience in welfare points (WPs), whose scale is too narrow to evaluate interventions decreasing suffering that is much more intense than the practically worst life.
  • Included an adjustment for “intense suffering” to mitigate the above shortcoming, but it only increased cost-effectiveness by 0.813 %, which I suspect is less than AIM intended.
• AIM moved from WPs to suffering-adjusted days (SADs) in 2024, but I believe the new system still dramatically underestimates the intensity of excruciating pain, and therefore the cost-effectiveness of interventions decreasing it. I calculate the past cost-effectiveness of the Shrimp Welfare Project (SWP) is 0.0757 % as high for AIM’s pain intensities as for mine.
• I strongly recommend AIM reconsiders their plan to incubate in early 2025 an organisation working on East Asian fish welfare. AIM has recommended this since 2022, but I estimate it is only 0.654 % as cost-effective as cage-free campaigns.
• I estimate the cost-effectiveness of Fish Welfare Initiative’s (FWI’s) farm program from January to September 2024 was 0.0711 DALY/$, or:
  • 7.16 times the cost-effectiveness of GiveWell’s top charities.
  • 4.26 % of the cost-effectiveness of broiler welfare campaigns.
  • 1.55 % of the cost-effectiveness of cage-free campaigns.
  • 0.0111 % of the past cost-effectiveness of SWP.
• Innovations for Poverty Action (IPA) did a cost-effectiveness estimation of FWI’s farm program commissioned by AIM which was last updated in March 2024. I estimate the mean cost-effectiveness they got ranges from 71.3 % to 92.7 % of mine. So there is remarkable agreement with my analysis.
• My results’ suggestion that FWI’s farm program is way less cost-effective than corporate campaigns for chicken welfare is in agreement with my analysis of AIM’s fish welfare recommendations. In particular, it is in line with my updates to AIM’s analysis of improving water quality on fish farms, which suggested this was 0.244 % (3.52 %) as cost-effective as cage-free campaigns at the time of recommendation paying (or not) for aeration equipment.
• I wonder whether AIM would have incubated FWI if they had accounted for both the probability of sentience and welfare range conditional on sentience. Rethink Priorities’ median welfare ranges were not available in 2019, when AIM recommended and incubated FWI, but I think AIM could still have done better than assuming fish and chickens have the same capacity for welfare.
• FWI says “most additional funding right now supports our R&D [research and development] work [not their farm program], which will enable us to become more cost-effective in the future”. Moreover, FWI highlights its future potential as the best reason for supporting them. I can see donating to FWI being better than to corporate campaigns for chicken welfare because of FWI’s current work having a higher ratio between future and present benefits. However, I think donating to SWP is clearly better than both options. I think they have the advantage of having been hugely more cost-effective than FWI, and also have significant future potential.

Fish welfare interventions recommended by Ambitious Impact

Context

The welfare range of the animals of a given species is equal to the product between their probability of sentience and welfare range conditional on sentience. Thanks to Michael St. Jules, I realised AIM implicitly assumed all species have welfare ranges conditional on sentience of 1, equal to that of humans, before they moved from assessing the benefits of animal welfare interventions in terms of WPs to SADs in 2024. Consequently, AIM had been overestimating the welfare range of species with lower welfare ranges conditional on sentience, and therefore the cost-effectiveness of helping animals of such species. For example, I believe helping fish was favoured relative to helping chickens, as fish have a lower welfare range conditional on sentience than chickens.

Calculations

Using RP’s median welfare ranges instead of AIM’s past assumption of welfare ranges conditional on sentience of 1, the 3 fish welfare interventions AIM has recommended have a significantly lower cost-effectiveness than the 4.59 DALY/$ I estimated for cage-free campaigns:

• Influencing European Union fish welfare policy through strategic work in Greece advocating for electrical stunning before slaughter, which was recommended in 2023, and incubated in 2024 via CAP, has a cost-effectiveness of 0.565 DALY/$ (= 334/0.526*0.089*0.01), 12.3 % (= 0.565/4.59) of my estimate for cage-free campaigns. I obtained the cost-effectiveness multiplying:
  • AIM’s estimate of 334 WPs/$.
  • The reciprocal of the probability of sentience AIM used of 52.6 %, such that this is not double-counted below.
  • RP’s median welfare range of carp of 0.089. Their value for salmon is 0.056, but I guess carp and salmon have similar welfare ranges, and that salmon’s lower value results from it having been less researched.
  • 0.01 DALYs averted per WP (= 1/100), since a practically maximally happy human life has an annual welfare of 100 WPs.
• East Asian fish welfare, which has been recommended from 2022 to 2024, and is scheduled for incubation in early 2025, has a cost-effectiveness of 0.0300 DALY/$ (= 24.4/0.725*0.089*0.01), 0.654 % (= 0.0300/4.59) of my estimate for cage-free campaigns. I determined the cost-effectiveness using the methodology above.
• Improving water quality on fish farms, which was recommended and incubated in 2019 via FWI, has a cost-effectiveness paying or not for aeration equipment of 0.0685 (= 1/0.013*0.089*0.01) and 0.989 DALY/$ (= 1/(9*10^-4)*0.089*0.01), 1.49 % (= 0.0685/4.59) and 21.5 % (= 0.989/4.59) of my estimate for cage-free campaigns. I obtained the cost-effectiveness using a methodology similar to the above, but using the reciprocal of estimates from AIM in $/WP (“$0.0009 - $0.013 per welfare point”), and excluding the probability of sentience because this is not integrated in the estimates of the WPs (although this is the only cost-effectiveness analysis I know from AIM where the estimates in WPs were not adjusted for sentience).

I wanted to link to the sheet with AIM’s cost-effectiveness estimates about improving water quality on fish farms so that people could check them as easily as possible. However, Vicky Cox, senior animal welfare researcher at AIM, said they do not want to make such an old analysis public. Feel free to ask Vicky for the sheet.

Discussion

Influencing European Union fish welfare policy through strategic work in Greece advocating for electrical stunning before slaughter

I conclude this is 12.3 % as cost-effective as cage-free campaigns. So I wonder whether AIM should have recommended it in 2023, and incubated it in 2024 via CAP. AIM could have relied on RP’s median welfare ranges. They were published in January 2023, before AIM’s in-depth report on the intervention was published in August 2023, and well before CAP’s incubation was announced in May 2024.

Nevertheless, I am actually glad AIM incubated CAP, as I consider they greatly underestimated how much their humane slaughter intervention decreases the time in pain weighted by intensity (not welfare range) per fish helped. I estimated CAP’s intervention is 12.3 % as cost-effective as cage-free campaigns using RP’s median welfare ranges, which I like. Yet, my result also depends on AIM’s estimate of 1.60 WPs gained per fish helped conditional on a welfare range of 1, which I think greatly underestimates the benefits. For my estimate of the DALYs averted per WP, it amounts to averting 0.0160 DALYs per fish helped (= 1.60*0.01) conditional on a welfare range of 1. In comparison, I estimate SWP’s Humane Slaughter Initiative, which also advocates for electrical stunning before slaughter, averts 1.37 DALYs per shrimp helped (= 0.0426/0.031) conditional on a welfare range of 1. As a result, if the reduction in the time in pain weighted by intensity per fish helped by CAP’s humane slaughter intervention matches that per shrimp helped by SWP’s, I would update towards CAP’s being 85.6 (= 1.37/0.0160) times as cost-effective as implied by my original correction to AIM’s estimate. That would make their intervention 10.5 (= 85.6*0.123) times as cost-effective as cage-free campaigns.

I think it is very easy to underestimate the benefits of interventions decreasing intense suffering, like advocating for humane slaughter, if they are estimated in WPs. Welfare per time only ranges from -100 to 100 WPs/year. This range may well be appropriate to assess lifetime welfare, and therefore interventions decreasing pain over a significant fraction of animals’ lives. However, it is too narrow to evaluate interventions decreasing intense suffering because the intensity of this is much greater than that of the practically worst life, whose welfare per time is just -100 WPs/year.

AIM included an adjustment for “intense suffering” in their cost-effectiveness analysis of CAP’s intervention to mitigate the above shortcoming of WPs. However, it has a negligible effect on the final cost-effectiveness. AIM’s “pessimistic” welfare gain is 10 times as large to account for “intense suffering”, bringing it to 0.042 WPs per fish helped conditional on a welfare range of 1. Nevertheless, AIM’s best guess of 1.60 is the mean between that pessimistic value, and an optimistic one of 3.17 which does not include the adjustment for “intense suffering”. Consequently, adjusting for “intense suffering” only increased cost-effectiveness by 0.813 % (= 1.60/((0.042/10 + 3.17)/2) - 1), which I suspect is less than AIM intended.

AIM moved from WPs to SADs in 2024, but I believe the new system still dramatically underestimates the intensity of excruciating pain, and therefore the cost-effectiveness of interventions decreasing it. I estimate the past cost-effectiveness of SWP is 639 DALY/$. For AIM’s pain intensities, and my guess that hurtful pain is as intense as fully healthy life^[1], I get 0.484 DALY/$, which is only 0.0757 % (= 0.484/639) of my estimate. Feel free to ask Vicky Cox, senior animal welfare researcher at AIM, for the sheet with their pain intensities, and the doc with my suggestions for improvement.

East Asian fish welfare

I strongly recommend AIM reconsiders their plan to incubate in early 2025 an organisation working on East Asian fish welfare. AIM has recommended this since 2022, but I estimate it is only 0.654 % as cost-effective as cage-free campaigns using RP’s median welfare ranges instead of AIM’s past assumption described above.

Improving water quality on fish farms

My estimate for the cost-effectiveness of improving water quality on fish farms without paying for aeration is the only one which is more than 10 % of my estimate for cage-free campaigns. Yet, I assumed these to be 1/5 as cost-effective as suggested by Saulius Šimčikas’ estimates published in 2019, when improving water quality on fish farms was recommended. So I calculate this was then 0.298 % (= 0.0149/5) and 4.30 % (= 0.215/5) as cost-effective as cage-free campaigns paying or not for aeration equipment

Fish Welfare Initiative’s farm program

Context

After noting the fish welfare interventions AIM has recommended look way less cost-effective than cage-free campaigns as assessed before implementation, I got interested in estimating FWI’s cost-effectiveness to see if that remains the case in practice. In addition, I had noted that Animal Charity Evaluators (ACE) had dropped their recommendation of FWI in 2024, although disclaiming:

Our [ACE’s] opinion is not that FWI’s work is less effective now than when ACE first decided to recommend them in 2022. In fact, we think their work has become more promising since then, but our bar for recommendation has changed as our evaluation methods have evolved.

Calculations

I estimate the cost-effectiveness of FWI’s farm program from January to September 2024 was 0.0711 DALY/$ multiplying:

• 5.42 fish-years improved per $. I determine this multiplying:
  • 3.08 fish-years improved per $ in 2023. I compute this multiplying 1 fish-year per 365.25 days by the following estimates from ACE’s last cost-effectiveness analysis of FWI (which FWI shared on their blog):
    • 126 M fish-days improved per $ by FWI’s farm program in 2023.
    • 112 k$ spent by FWI’s farm program in 2023.
  • 1.76 fish defined by FWI as helped per $ from January to September 2024 as a fraction of those in 2023. I compute this multiplying:
    • 7.10 fish defined by FWI as helped per $ from January to September 2024, as reported by FWI.
    • 4.04 fish defined by FWI as helped per $ in 2023, which is the ratio between 452 k fish defined by FWI as helped in 2023, as reported by FWI, and 112 k$ spent by FWI’s farm program in 2023, based on ACE’s cost-effectiveness analysis.
• “14.75” WPs gained per fish-year improved conditional on a welfare range of 1, as estimated by Ambitious Impact (AIM) for improvements in water quality related to ensuring adequate oxygen levels. I again wanted to link to the sheet with AIM’s estimate, but Vicky said they do not want to make such an old analysis public. You can ask Vicky for the sheet.
• 0.089 for the welfare range, which is RP’s median welfare range of carp. Haven clarified FWI helps carp.
• 0.01 DALYs averted per WP, since a practically maximally happy human life has an annual welfare of 100 WPs.

Haven King-Nobles, FWI’s co-founder and executive director, noted the estimates for the fish-years improved per $, and WPs gained per fish-year improved conditional on a welfare range of 1 are speculative right now, as FWI is still learning. At the same time, Haven believes they are as speculative as most similar numbers of other animal welfare interventions.

My estimated cost-effectiveness is:

• 7.16 times the cost-effectiveness of GiveWell’s top charities.
• 4.26 % of the cost-effectiveness of broiler welfare campaigns.
• 1.55 % of the cost-effectiveness of cage-free campaigns.
• 0.0111 % of the past cost-effectiveness of SWP.

Discussion

I would have preferred to estimate the cost-effectiveness of FWI’s farm program based on the decrease in the time fish spend in pain, as I did for corporate campaigns for chicken welfare and SWP, for greater accuracy and comparability. The Welfare Footprint Project (WFP) is working on estimating the time farmed fish spend in pain. Haven commented FWI aspires to assess their programs based on WFP’s methodology.

IPA conducted a (time-constrained) cost-effectiveness estimation^[2] of FWI’s farm program as projected for the 10th year of FWI’s operations, 2028 (= 2019 + 10 - 1). The analysis was commissioned by AIM, and was last updated in March 2024. Among “~10” Monte Carlo simulations with “1,000 trials” each, IPA got a mean cost-effectiveness ranging from 57 to 74 WPs/$ conditional on a welfare range of 1^[3]. For my estimates of the welfare range, and DALYs averted per WP, the mean cost-effectiveness estimated by IPA ranges from 0.0507 (= 57*0.089*0.01) to 0.0659 DALY/$ (= 74*0.089*0.01), or 71.3 % (= 0.0507/0.0711) to 92.7 % (= 0.0659/0.0711) of what I got. So there is remarkable agreement with my analysis.

My results’ suggestion that FWI’s farm program is way less cost-effective than corporate campaigns for chicken welfare is in agreement with my analysis of AIM’s fish welfare recommendations. In particular, it is in line with my updates to AIM’s analysis of improving water quality on fish farms, which suggested this was 0.298 % (4.30 %) as cost-effective as cage-free campaigns at the time of recommendation paying (or not) for aeration equipment. FWI ended up not promoting aeration due to this incentivising farmers to stock fish more densely, thus constraining the welfare improvements^[4]. Nonetheless, I assume AIM’s case for incubating FWI was based on advocating for aeration being cost-effective in its own right, or FWI pivoting towards a similarly cost-effective intervention, not on FWI eventually finding much more cost-effective interventions competitive with cage-free campaigns.

I wonder whether AIM would have incubated FWI if they had accounted for both the probability of sentience and welfare range conditional on sentience. Rethink Priorities’ median welfare ranges were not available in 2019, when AIM recommended and incubated FWI, but I think AIM could still have done better than assuming fish and chickens have the same capacity for welfare:

• The ratio between RP’s median welfare ranges of carp and chickens is 26.8 % (= 0.089/0.332).
• AIM would have obtained a value of 21.3 % (= (1*0.0455)^0.5), 79.5 % (= 0.213/0.268) of the above, from the geometric mean between 2 arguably plausible estimates available in 2019:
  • 1, as implied by AIM not accounting for the probability of sentience and welfare range conditional on sentience in their cost-effectiveness analysis of improving water quality on fish farms.
  • 4.55 % (= 10*10^6/(220*10^6)), as inferred from the ratio between the number of neurons of the zebrafish and red junglefowl.

FWI says “most additional funding right now supports our R&D work [not their farm program], which will enable us to become more cost-effective in the future”. According to FWI’s budget for 2025, which I am glad is public, R&D will account for 35.0 % (= 299*10^3/(854*10^3)) of their spending outside operations, and their farm program for 20.7 % (= 177*10^3/(854*10^3)). So, if FWI thought the marginal cost-effectiveness of their R&D work was higher than that of their farm program, they could move funds from the latter to the former until their marginal cost-effectiveness became the same. There would be limited room to do this if the farm program was supported by restricted funding, but Haven clarified this is not the case. Haven endorsed the current allocation of funds, explaining their farm program helps build the fish welfare ecosystem, and inform their R&D work. If there are no significant differences in marginal cost-effectiveness across programs, “most” of FWI’s additional funding going to R&D makes sense if the other programs have steeper diminishing returns, i.e. if their marginal cost-effectiveness decreases faster with funding. I think all of this is plausible, so I do not have any clear disagreement with FWI’s budget for 2025.

There is a list with the best arguments for donating to FWI on their donation page. The top 2 are^[5]:

1. FWI’s future potential for impact: About 67% of our current budget (specifically our R&D, exploratory programs, and China budget items) goes towards developing more cost-effective interventions in the future rather than having a direct impact. We conduct this intervention research in what we believe is an unusually rigorous and ground-proofed way. For examples, see our recent studies focused on developing interventions on satellite imagery and feed fortification.
2. FWI’s current impact: We currently estimate that we’ve improved the lives of over 2 million fishes. This makes FWI one of the most promising avenues in the world to reduce farmed fish suffering, and likely the most promising avenue in the world to reduce the suffering of farmed Indian major carp, one of the largest and most neglected species groups of farmed fishes.

I agree with this ordering. I estimated FWI’s farm program is way less cost-effective than corporate campaigns for chicken welfare and SWP, but I neglected the future benefits of FWI’s present work, including the value of information. On the other hand, these would have to be large multiples of the direct benefits for FWI to be as cost-effective as those interventions, even neglecting their own future benefits. For FWI to be as cost-effective as:

• Broiler welfare campaigns, 23.5 (= 1/0.0426) times.
• Cage-free campaigns, 64.5 (= 1/0.0155) times.
• SWP has been, 9.01 k (= 1/(1.11*10^-4)) times.

FWI guesses satellite imaging can increase the fraction of visits to farms which result in improved conditions from 5 % to something from 50 % to 99 %. If so, and the cost of FWI’s farm program was proportional to the number of visits to farms, satellite imaging would make it 14.9 (= (0.5 + 0.99)/2/0.05) times as cost-effective. In reality, FWI’s farm program has other costs, so the cost-effectiveness would not become so large. Even if it did, ignoring the future benefits of corporate campaigns and SWP, it would become:

• 107 times as cost-effective as GiveWell’s top charities.
• 63.5 % as cost-effective as broiler welfare campaigns.
• 23.1 % as cost-effective as cage-free campaigns.
• 0.166 % as cost-effective as SWP.

I can see donating to FWI being better than to corporate campaigns for chicken welfare because of FWI’s current work having a higher ratio between future and present benefits. However, I think donating to SWP is clearly better than both options. I think they have the advantage of having been hugely more cost-effective than FWI, and also have significant future potential. Aaron Boddy, SWP’s chief operations officer, commented:

We’re having our annual team retreat (which we call “Shrimposium”) next week, during which we hope to map out how we can deploy stunners in such a way as to catalyse a tipping point so that pre-slaughter stunning becomes the industry standard.

We’ve had some good indications recently that HSI [SWP’s Humane Slaughter Initiative] does contribute to “locking-in” industry adoption, with Tesco and Sainsbury’s recently publishing welfare policies, building on similar wins in the past (such as M&S and Albert Heijn).

This has always been the Theory of Change for the HSI project. Although we’re very excited by how cost-effective it is in its own right, ultimately we want to catalyse industry-wide adoption - deploying stunners to the early adopters in order to build towards a tipping point that achieves critical mass. In other words, over the next few years we want to take the HSI program from Growth to Scale.

Acknowledgements

Thanks to Filip Murár, Haven King-Nobles, Jennifer-Justine Kirsch, Katrina Sill, Thomas Billington, and Vicky Cox for feedback on the draft^[6].

1. ^{^}

    This means 1 h of hurtful pain plus 1 h of fully healthy life is neutral.

2. ^{^}

    IPA only uses the term “cost-effectiveness analysis” for estimates which involve more credible sources for the benefits and cost.

3. ^{^}

    This is not made explicit in IPA’s analysis. However, their point estimate of 18 WPs gained per fish-year improved is the mean between AIM’s estimates of 14.75 and 21.5 which are conditional on a welfare range of 1. I again wanted to link to the cells in AIM’s sheet with these values, but Vicky said they do not want to make such an old analysis public. You can ask Vicky for the sheet.

4. ^{^}

    Haven commented:

   FWI decided pretty early on that paying for aeration equipment would probably never be a significant strategy of ours. The reason for that is that it seems that providing aeration to farms would mean that farmers ended up intensifying (e.g. increasing the stocking density), and thus potentially causing more suffering. This was illustrated to me in my first visit to India, when I asked a farmer how he’d feel if my non-profit were to buy him aerators for his farm. “Great!” he said. “Then I could farm twice as many fish!”.

5. ^{^}

    Vicky commented a key aspect of FWI’s impact is “figuring out how to improve the welfare of fish farmed by poorer farmers in informal markets which make up a large % of the fish market. Currently we don’t know what works there, and so there is lots of suffering we don’t know how to avert”. Filip Murár, research analyst at AIM, commented the value of information is potentially very high. I assume FWI were categorising that under “FWI’s future potential for impact”.

6. ^{^}

    I listed the names alphabetically.