Corporate campaigns for chicken welfare are 10,000 times as effective as GiveWell's Maximum Impact Fund?

By Vasco Grilo🔸 @ 2022-07-28T08:22 (+78)

Summary

• This analysis estimates the cost-effectiveness of corporate campaigns for chicken welfare (CCCW), and compares it with that of GiveWell's Maximum Impact Fund (MIF) (see Methodology).
• The results were obtained with this Colab, and the key ones are summarised in the table below (for more, see Results). Comments about how to interpret them are welcome.
• The ratio between the mean cost-effectiveness of CCCW and MIF is estimated to be 10 k (i.e. 10^4).
  • The specific value depends on very uncertain parameters, such as the ones defining the badness of the conditions of factory-farmed chickens relative to the goodness of fully healthy chicken life, and the moral weight of chickens relative to humans.
  • However, it appears to be robustly larger than 1 (see Discussion).

Acknowledgements

Thanks to Cynthia Schuck-Paim, Kieran Greig, Lewis Bollard, and Saulius Simcikas.

Introduction

About 2 year ago Stephen Clare and Aidan Goth (S&A) published an analysis comparing the cost-effectiveness of The Humane League^[1] (THL) and Against Malaria Foundation (AMF). The analysis presented here has the same motivation of better allocating resources between different causes (namely, animal welfare and global health and development). In addition, it relies on data from the Welfare Footprint Project (WFP; overviewed here) to model more accurately the relative improvement in the conditions of broilers and laying hens^[2].

Methodology

The ratios between the cost-effectiveness statistics of CCCW and MIF were calculated via determining the respective cost-effectiveness distributions via a Monte Carlo simulation in this Colab. All distributions were assumed to be independent, and their parameters were estimated from quantiles as explained here.

The methodology to determine the cost-effectiveness of CCCW and MIF, and assess room for more funding are presented below.

Cost-effectiveness of corporate campaigns for chicken welfare

The cost-effectiveness of CCCW was based on its cage-free campaigns, and determined from the product between:

• The cost of improving the conditions of chickens^[3] (cyear/$).
• The improvement in the conditions of chickens^[4] (cQALY/cyear).
• The moral weight of chickens relative to humans^[5] (QALY/cQALY).

Cost of improving the conditions of chickens

The cost of improving the conditions of chickens was supposed to follow a lognormal distribution with 5th and 95th percentiles in cyear/$ equal to:

• For broilers, 0.2 and 90.
• For laying hens, 12 and 160.
• For both combined, 9.6 and 120.

These are the lower and upper bound of the 90 % subjective confidence interval estimated for the cost-effectiveness of "broiler and cage-free campaigns" in this analysis from Saulius Simcikas^[6]. Potential concerns with those estimates are discussed here.

Improvement in the conditions of chickens

The improvement in the conditions of chickens were calculated assuming weights of:

• 12 % for changing broilers from a conventional scenario (CS) to a reformed scenario^[7] (RS).
• 88 % for changing laying hens from conventional cages (CC) to cage-free aviaries^[8] (CFA).

The fractions above were computed with the Guesstimate model of Saulius' analysis, and are the ratios between the chicken years affected by campaigns for:

• Numerator: either broilers or laying hens.
• Denominator: broilers and laying hens.

The conditions of chickens were determined from the ratio between:

• The sum of the products between the time experiencing and utility of each type of pain^[9]:
  • The time experiencing each type of pain was calculated based on data from the WFP.
  • The utility of each type of pain was defined based on my intuition and the description of the 4 types of pain presented in Box 1.2 of The Comparative Measurement of Animal Welfare: the Cumulative Pain Framework.
• The lifespan:
  • For broilers, 42 d^[10].
  • For laying hens, 72 weeks^[11]:
    • Hatchery phase during week 1.
    • Rearing phase during weeks 2 to 18.
    • Laying phase during weeks 19 to 72.

Broilers' time experiencing each type of pain

The time experiencing each type of pain was defined as a lognormal distribution with 5th and 95th percentiles in h equal to the lower and upper bound of the 90 % confidence interval presented in this page^[12] (see this):

• For excruciating pain:
  • For CS, 8.830/3600 and 51.57/3600.
  • For RS, 2.100/3600 and 9.06/3600.
• For disabling pain:
  • For CS, 33.01 and 67.53.
  • For RS, 6.20 and 28.32.
• For hurtful pain:
  • For CS, 195.08 and 472.12.
  • For RS, 125.46 and 385.94.
• For annoying pain:
  • For CS, 212.82 and 436.52.
  • For RS, 188.66 and 489.96.

Laying hens' time experiencing each type of pain

The time experiencing each type of pain was defined as follows:

• For the hatchery phase^[13]:
  • Disabling pain for a lognormal distribution with 5th and 95th percentiles equal to 5 % and 15 % of the time spent awake during this phase^[14].
  • Hurtful pain for the rest of the duration of the phase.
• For the rearing phase^[15]:
  • For CC:
    • Hurtful pain for a lognormal distribution with 5th and 95th percentiles equal to 20 % and 50 % of the time spent awake during this phase.
    • Annoying pain for the rest of the duration of the phase.
  • For CFA, neutral utility.
• For the laying phase, a lognormal distribution with 2.5th and 97.5th percentiles in h equal to the lower and upper bound of the 95 % confidence interval presented in this page of the WFP^[12]:
  • For excrutiating pain:
    • For CC, 0.03 and 0.07.
    • For CFA, 0.02 and 0.06.
  • For disabling pain:
    • For CC, 354.1 and 508.8.
    • For CFA, 100.2 and 211.9.
  • For hurtful pain:
    • For CC, 3,265 and 4,844.
    • For CFA, 991 and 2,492.
  • For annoying pain:
    • For CC, 5,615 and 7,827.
    • For CFA, 1,197 and 2,956.

Negative utility of each type of pain

The negative utility of each type of pain in pQALY/pyear was set to a lognormal distribution with 5th and 95th percentiles equal to^[16]:

• For excruciating pain: 100 and 10 k.
• For disabling pain: 1 and 100.
• For hurtful pain: 0.1 and 10.
• For annoying pain: 0.01 and 1.

Moral weight of chickens relative to humans

The moral weight of chickens relative to humans was determined from the product between:

• The probability of chickens being moral patients, as defined by Luke Muehlhauser here, which was set to 80 % according to this section of Open Philanthropy's (OP's) 2017 Report on Consciousness and Moral Patienthood.
• A loguniform distribution whose 10th and 90th percentiles in QALY/cQALY were set to 5*10^-5 and 10. These are the lower and upper bounds of the "80 % prediction interval" guessed by Luke Muehlhauser here for the moral weight of chickens relative to humans conditional on the former being moral patients^[17] (see "Moral weights of various species").

The distribution for the moral weight of poultry might depend on the theory of consciousness. The above product is implicitly assumed to represent the weighted mean of the moral weight distributions of the various theories of consciousness. These are, in turn, supposed to produce (summable) moral weight distributions in QALY/cQALY.

A comparison between the implications of the above and those of the Weighted Animal Welfare Index of Charity Entrepeneurship is avaible here. Potential concerns about calculating expected moral weights are discussed here.

Cost-effectiveness of GiveWell's Maximum Impact Fund

The cost-effectiveness of MIF was determined from the product between:

• The funding bar of MIF as a multiple of the cost-effectiveness of GiveDirectly^[18], which was set to 10 in agreement with GiveWell's best guess for its near term cost-effectiveness bar^[19].
• The cost-effectiveness of GiveDirectly, which was estimated from the ratio between:
  • The utility of increasing consumption, which was set to 0.5 QALY per doubling of consumption in line with Open Philanthropy's moral weights^[20].
  • The cost-benefit ratio of GiveDirectly, which was set to a lognormal distribution with 2.5th and 97.5th percentiles equal to 130.56 $ and 1,185.44 $ per doubling of consumption. These quantiles were estimated here by Sam Nolan.

Defining the 1st and 3rd factors as constants underesimates the uncertainty of the cost-effectiveness of MIF. However, the major driver for the uncertainty of the ratio between the cost-effectiveness of CCCW and MIF is the cost-effectiveness of CCCW.

Room for more funding of corporate campaigns for chicken welfare

OP's grants in the focus areas broiler chicken welfare and cage-free reforms were analysed to better understand whether CCCW have room for more funding. The 80 grants regarding these areas are in tab “Grants” of this Sheet^[21]. The grant amounts were adjusted for inflation based on data from in2013dollars (see tab “Inflation”), and are expressed in 2020-$.

A linear regression of the annual amount granted on year was determined with this Colab.

Results

The results for the cost-effectiveness are presented below for a Monte Carlo simulation with 10 M samples. I encourage the readers to make a copy of the Colab model, and select their preferred parameters (e.g. those which define the moral weight). The model is fully commented (the inputs section is at the top), and could be run in 30 s for 10 M samples.

Cost-effectiveness of corporate campaigns for chicken welfare

Cost-effectiveness of GiveWell's Maximum Impact Fund

Ratios between the cost-effectiveness statistics of corporate campaigns for chicken welfare and GiveWell's Maximum Impact Fund

Room for more funding of corporate campaigns for chicken welfare

The results for the linear regression of the annual amount granted by OP to CCCW on year, between 2016 and 2021, are presented below.

Discussion

Improvement in the conditions of chickens

To find good heuristics for the mean relative improvement in the conditions of broilers and laying hens, these can be compared to the relative reduction in the median time experiencing each type of pain. According to the WFP:

• For broilers (whose conditions improved by 45.4 %):
  • For excruciating pain, 81.5 % (from 30.20 s to 5.58 s).
  • For disabling pain, 65.7 % (from 50.27 h to 17.26 h).
  • For hurtful pain, 23.4 % (from 333.60 h to 255.70 h).
  • For annoying pain, -4.51 % (from 324.67 h to 339.31 h).
• For laying hens (whose conditions improved by 64.6 %):
  • For excruciating pain, 20 % (from 0.05 h to 0.04 h).
  • For disabling pain, 63.8 % (from 431.41 h to 156.04 h).
  • For hurtful pain, 57.0 % (from 4,054.47 h to 1,741.92 h).
  • For annoying pain, 69.1 % (from 6,721.17 h to 2,076.79 h).

Consequently, for the negative utility of each type of pain defined at the end of this section:

• The relative reduction in the median time broilers experience disabling pain (65.7 %) is a decent heuristic for the mean relative improvement in broilers' conditions (45.4 %).
• The relative reduction in the median time broilers experience disabling, hurtful, or annoying pain (63.8 %, 57.0 %, or 69.1 %) are good heuristics for the mean relative improvement in hens' conditions (64.6 %).

Moral weight of chickens relative to humans

The mean moral weight of chickens relative to humans of 2 QALY/cQALY may appear too high. Nevertheless:

• Based on this analysis, using other distributions for the moral weight of chickens relative to humans conditional of the former being moral patients would lead to a higher mean.
• The 90th percentile of the moral weight of chickens relative to humans of 8 QALY/cQALY (= 0.8*10) implies the mean is larger than 0.8 QALY/cQALY^[23] (= (1 - 0.9)*8).

In other words, for the mean moral weight of chickens relative to humans to be much smaller than 1, one has to be very confident that it cannot exceed 1. For example, for a mean of 0.01, one would have to assign a probability smaller than 1 % (= 1/0.01) to the moral weight being larger than 1. However, this is arguably hard to defend given the current incomplete understanding of consciousness.

Cost-effectiveness of CCCW

It is worth noting that:

• The mean cost-effectiveness of CCCW is of the order of 10^2 QALY/$, which means 0.01 $ leads to the equivalent of 1 year of fully healthy human life.
• The mean cost-effectiveness of improving the conditions of laying hens is similar to the 200 QALY/$ estimated by Michael Dickens here for cage-free campaigns (see results).
• The mean cost-effectiveness of CCCW for laying hens is 3 (= 204/64.4) times as large as that of broilers.

Ratio between the mean cost-effectiveness of corporate campaigns for chicken welfare and GiveWell's Maximum Impact Fund

The ratio between the mean cost-effectiveness of CCCW and MIF is 10 k. This is similar to the value of 20 k (= 204.698/0.00970468) of the ratio between the mean direct effects of cage-free campaigns and AMF estimated by Michael Dickens here (see results).

I suspect disagreements about the ratio between the mean cost-effectiveness of CCCW and MIF are primarily driven by different intuitions respecting^[24]:

• The badness of the conditions of CS, RS, CC and CFA relative to the goodness of fully healthy chicken life.
• The shape and top percentiles of the distribution of the moral weight of chickens relative to humans.

Nonetheless, I believe most informed choices for these factors result in a ratio between the mean cost-effectiveness of CCCW and MIF which is much larger than 1^[25]. For example, the mean product between the conditions of laying hens in CC and moral weight of chickens relative to humans does not differ much from those of S&A and Charity Entrepeneurship (CE). It is 2 times the one of S&A, and 9 times the one implied by CE's  Weighted Animal Welfare Index (WAWI):

• For this analysis, it is 5.3 -QALY/cyear^[26].
• For S&A, it is 2.7 -QALY/cyear^[26].
• For WAWI, it is 0.40 -QALY/cyear^[27].

As a consequence, the ratio between the mean cost-effectiveness of CCCW and MIF (12 k) is 1 k times as large as the ratio between the largest and smallest of the values above (5.28/0.399 = 13). This ratio is equal to the product between 4.78 k and the mean moral weight of chickens relative to humans if these are moral patients, which would have to be 2*10^-4 (= 1/(4.78 k)) for the ratio to be 1.

That being said, such ratio being 10 k does not mean donating to CCCW is 10 k times as good as to MIF. The cost-effectiveness estimates do not account for all effects^[28], and such large differences in true cost-effectiveness are arguably unusual. On the other hand, the ratio being much larger than 1 suggests CCCW are better than MIF, even if cost-effectiveness estimates can be misleading.

Room for more funding of corporate campaigns for chicken welfare

The annual amount granted by OP to CCCW was 9.68 M$ in 2021, and has decreased 1 M$/year between 2015 and 2021. However, this downwards trend is not significant, given the p-value of 40 % for the null hypothesis of null slope.

In addition, according to Kieran Greig, who surveyed groups working on CCCW globally:

These campaigns have some pretty significant room for more funding. Easily in the millions of dollars per year.^[29]

Moreover, Saulius recently estimated that "in 2019-2020 chicken welfare reforms affected 65 years of chicken life per dollar spent". Using this estimate would increase the cost-effectiveness of CCCW by 30 % (= 65/45.6 - 1).

For these reasons, the ratio between the mean cost-effectiveness of CCCW and MIF would hardly decrease to 1 due to considerations regarding room for more funding.

1. ^{^}

   THL has been an ACE top charity since 2012.

2. ^{^}

   This aspect is noted by Stephen and Aidan in the future work section of their analysis (for laying hens):

   - "However, a better understanding of how bad aviaries are relative to battery cages could be valuable".

3. ^{^}

   A cost of improving the conditions of laying hens of 1 cyear/$ means 1 $ funds improved conditions for 1 year of chicken life.

4. ^{^}

   An improvement in the conditions of laying hens of 1 cQALY/cyear is as valuable as 1 year of fully healthy chicken life. As suggested here, fully healthy chicken life could be interpreted "as [a chicken] living with all needs met, no or minimal fear of predation and disease-free (e.g. perhaps the best moments on a very good farm animal sanctuary)".

5. ^{^}

   A moral weight of chickens relative to humans of 1 QALY/cQALY means 1 year of fully healthy human life is as valuable as 1 year of fully healthy chicken life.

6. ^{^}

   See bottom row of the top table.

7. ^{^}

   According to section "Conventional and Reformed Scenarios" of Chapter 1 of Quantifying pain in broiler chickens, the "conventional scenario" is represented by the use of "fast-growing breeds", and the "reformed scenario" by "a slower-growing strain".

8. ^{^}

   This assumption is discussed here.

9. ^{^}

   It is possible to experience multiple types of pain simultaneously. All pain-free time is assumed to be morally neutral.

10. ^{^}

    In agreement with section "Conventional and Reformed Scenarios" of Chapter 1 of Quantifying pain in broiler chickens.

11. ^{^}

    The typical duration of each phase was provided by Cynthia Schuck-Paim.

12. ^{^}

    Search for "Total Time in Pain".

13. ^{^}

    According to Cynthia Schuck-Paim, for the hatchery phase:

    - "Evidence indicates that female chicks experience important levels of stress during commercial hatchery processing (beak trimming being possibly the most painful, with pain lasting days) and transport to rearing facilities, which have several behavioral and hormonal effects later in life. Null utility in the first week of life is unlikely (we [Welfare Footprint Project] have not measured it, but my personal guess is that welfare is net negative regardless of the housing system)".

    - "Some fraction [of the time may be allocated] to disabling pain to account for beak trimming (e.g. chicks have been observed to be unwilling to eat or drink for over 24 hours after the procedure) and the high levels of stress (and some injury) during transport. Something around 5-15% of this first week".

14. ^{^}

    The time spent awake is set to 16 h/d. According to Chapter 3 of Quantifying Pain in Laying Hens, "only hours spent awake (16 hours per day) are considered (e.g., if pain is estimated to last 1-2 days, this translates into 16-32 hours of pain)".

15. ^{^}

    According to Cynthia Schuck-Paim, for the rearing phase:

    - For caged hens:

       -- "Restriction of movement and natural behaviors (foraging, exploration, perching) and other forms of agency is already present if chicks are caged during rearing. Utility is possibly not null".

       -- "Some fraction [of the time may be allocated] to hurtful pain as well (something around 20-50%, as the deprivation to forage, move, explore, etc is unlikely to represent only an annoying experience; see The Burden of Psychological Pain in Laying Hens: Behavioral Deprivation)".

    - For cage-free hens:

       -- "Under good management, one could assume that the main welfare challenges are not yet present at this stage (neutrality could be justified)".

    - "Disabling pain, as a general rule, is unlikely since the main drivers of  pain at this intensity are not yet present".

16. ^{^}

    The percentiles were defined based on my intuition and the description of the 4 types of pain. The geometric mean of the defined percentiles implies null median utility for 1 h of fully healthy chicken life plus one of the below experiences:

    - 4 s of excruciating pain.

    - 6 min of disabling pain.

    - 1 h of hurtful pain.

    - 10 h of annoying pain.

17. ^{^}

    The reasons for selecting a loguniform distribution are described in the last paragraph of this analysis.

18. ^{^}

    Note this underestimates the uncertainty of the cost-effectiveness of MIF, as the cost-effectiveness of GiveDirectly is less uncertain than the other GiveWell's top charities.

19. ^{^}

    According to this article from GiveWell, "In the near term we [GiveWell] will:

    - "Increase our cost-effectiveness bar to 10x cash (current best guess, subject to change)".

20. ^{^}

    "Our [Open Philanthropy's] new approach values a DALY averted twice as highly, equal to a 2-unit (rather than 1-unit) increase in the natural log of any individual’s income. (This is equivalent to increasing 200 people’s incomes by 1% – i.e., in our favored units, equal to $100K in units of marginal dollars to individuals making $50K.)".

21. ^{^}

    The focus areas of the downloadable CSV file of OP's grants database are in agreement with those of the website.

22. ^{^}

    Considering null slope as the null hypothesis.

23. ^{^}

    I owe this remark to Guy Raveh's comment.

24. ^{^}

    I tend to believe disagreements about the relative improvement in the conditions of laying hens when these change from CC to CFA have been atenuated by WFP's research.

25. ^{^}

    This is in line with what is mentioned in the Summary of S&A:

    - "In this model, in most of the most plausible scenarios, THL appears better than AMF".

26. ^{^}

    Calculated from the product between the mean conditions of laying hens in CC and the mean moral weight of chickens relative to humans, as these were modelled as independent.

27. ^{^}

    Calculated from the negative of the product between:

    - Probability of "USA FF [factory-farmed] laying hens (battery cages)" feeling pain of 70 %.

    - The "total welfare score (with evidence)" (TWS) of "USA FF laying hens (battery cages)" of -57.

    - The reciprocal of the maximum TWS of 100, which was supposed equivalent to 1 QALY if experienced for 1 year.

28. ^{^}

    In terms of CCCW, potential concerns with the estimates for the cost to improve conditions are discussed here. For global health and development, and therefore MIF, there is the meat-eater problem (quantified here) and unclear effects on population size, among others.

29. ^{^}

    Further details are confidential:

    - "I apologize that I can't share too much specifically as I promised organizations that those results would be confidential".