Summary: Using Animal Charity Evaluator’s figures, I estimate the amount donated to an effective animal charity to equal the harm caused by a typical American diet compared to veganism. This figure is surprisingly low: $2-5 per year. This suggests that personal dietary change, relative to other things we can do, is fairly ineffective. Yet most EAs interested in animal welfare are eager that others, including other EAs, stop using animal products. I explore a variety of means of resolving this tension, and recommend a large downward adjustment to the efficacy of animal charities is the best solution. 1
Introduction
Animal agriculture causes vast amounts of suffering: not only in virtue of animals being killed for their meat, but also the suffering they undergo whilst alive in industrial agriculture. Many in EA consider this an important and neglected cause, and among Effective Altruists who focus on animal agriculture, reducing consumption of animal products (mainly meat, eggs, and dairy) is cause commonly advocated for. Reduce the consumption of animal products, and you reduce the production of animal suffering. 2This activism is extended ‘intra’ EA, with advocacy for EAs individually and at corporate events to refrain from serving animal products.
This invites the question as to how valuable an individual changing their diet is: EAs might prefer to keep eating meat, and be keen on quantifying just how much harm they are doing. These estimates are tricky: both Katja Grace and Jeff Kaufman have offered approximate trade-offs for animal product consumption in terms of human welfare (or welfare generally), but a major uncertainty in their calculations is how to weigh human versus non-human welfare: are years of human life more valuable than years of chicken life? Or maybe animal lives in animal agriculture are further below zero than typical human lives are above it.
Happily, thanks to Animal Charity Evaluators, we have some estimates made of how cost-effectively one can avert animal suffering, which is denominated in a similar way to the expected costs of eating animal products derived by Brian Tomasik and Peter Hurford. We can use these figures to estimate the cost equivalent in terms of donations to an effective charity for use of animal products.
Calculation
Peter Hurford estimates the typical American diet to cause 5.5 years of animal suffering. Some components cause disproportionately more (e.g. eggs) or less (e.g. beef) than their proportion of weight or calorific content. Thus (borrowing Hurfords numbers) a vegetarian diet where one maintains consumption of milk and eggs but eschews meats causes around a third of the animal suffering of a fully carnivorous diet. If avoids eggs as well and only consumes dairy, this avoids 98.6% of the suffering. 3
How much would it cost to avert 5.5 years of animal suffering? Earlier suggestions put the price of averting one year of animal suffering into sub-dollar range. Probably the best contemporary estimates are from Animal Charity Evaluators, who offer estimates for donations to particular charities in terms of animal lives averted. The ‘raw figures’ in terms of animal lives saved are similarly optimistic: ACE estimates the Humane Leaguesaves 3.4 animal lives per dollar spent, and the other two organizations (Mercy for Animals and Animal Equality) do even better – 8.8 and 10.1 animals per dollar respectively.
ACE often use their leafleting and online ad calculators in their reviews of these charities. Helpfully, these output figures directly in terms of life-years averted. Let’s use the calculator for online ads. The central estimate is 2.64 years of factory farming averted per dollar (for leafleting, the figure is 2.02 years/$). The estimate for the ‘donation equivalent’ to a year of veganism is thus:
Donation equivalent of veganism = Animal suffering of diet/suffering averted per dollar $2.08/year = 5.5 years of suffering per year/2.64 years averted per dollar.
One can use the prior adjustments to work out the donation equivalent of dairy alone versus full veganism (lactovegetarianism):
Donation equivalent of lactovegetarinaism = $2.08/year * proportion of harm attributable to dairy (1/72.2) = $0.03 / year
Or eggs and dairy (ovolactovegetarianism) over full veganism:
Donation equivalent of O-Lveg = $2.08 * proportion eggs and dairy (~0.357) = $0.74/year
These calculations elide differences between species, and implictly demand judgement calls as to how to equate years ‘as a chicken’ versus ‘as a cow’, etc. We can use the calculator another way to avoid these difficulties: as it also computes the number of ads needed to make one person stop consuming animal products, and for how long that effect is expected to last. 4 The calculator dis-aggregates animal products into meat, eggs, and dairy. Fairly conservatively, we can take the lowest yield of these (dairy) to be the cost of an effective vegan, as once we have donated enough to make one person give up dairy, we have also made more than one person give up eggs and meat, and we can ignore the additional fractions of people avoiding eggs or meat alone. 5 The calculator suggests that for $100 one can expect 3.68 people to give up dairy for an average of 6.2 years. Thus the approximate cost per year of veganism is:
Cost per year of 'veganism' = $100/(3.68*6.2) years of 'veganism' = $4.38 /year
The ‘cost per year of lactovegetarianism’ or vegetarianism can be estimated by looking at the (better) conversion rates to get people to stop using eggs (7.37 and 11.05 respectively), giving the cost per year of lactovegetarianism and vegetarianism at $2.12/year and $1.45/year respectively.
Uncertainties and over-estimation
These figures seem remarkably low. Should you believe them?
There are a variety of minor considerations which could drive the estimate down somewhat. One may want to look at the balance of animal products in particular diets, or disagree with the implied intra-species comparisons. These might modify the bottom line figures somewhat, but not much – and, in any case, one could look at ‘cost per vegan-year’ estimates which obviate these sorts of worries. Another concern is you cannot ‘just’ buy ads, and the charities ACE recommends have some other activities – however, ACE does not believe these activities are always less cost effective than online ads, so correcting for other activities could get the estimate to go up instead of down.
The chief concern should be overestimation by ACE. Granting the estimates are unbiased, one should still expect regression to the mean. If the underlying distribution of ‘true’ charity effectiveness is lognormal (or worse), the degree of regression could be orders of magnitude.
Offsets and second-order Outcomes
Suppose you have a less regressive mindset, and you are confident the real figures are not-too-far from ACE’s estimates. One conclusion EAs might make is that their personal diets are no big deal, easily swamped as it is by the consequences of donations. Further, the very low cost of ‘buying’ veganism elsewhere makes it apt for moral offsetting 6 -most carnivores (EAs included) probably have a consumer surplus from animal products greater than a few dollars per year (equating to cents per meal). However, most EAs who are interested in animal welfare hesitate to endorse these conclusions, and are still eager for their fellows to become vegan notwithstanding the greater ease of doing an equivalent amount of good by donation. Why?
One possibility is second order effects, currently neglected, that make veganism much more valuable. Being vegan might persuade others to refrain from animal products, and perhaps pushing social norms to a ‘tipping point’ where there’s a dramatic change in our attitudes towards animals. Perhaps – someone advocating for animal welfare who nonetheless eats meat is unlikely to persuasive. Yet it seems unlikely these second order effects will be much greater than the primary one of the person involved (I’m confident my lifelong vegetarianism has ‘converted’ <<1 carnivore). 7 The small ‘cost’ of the direct offset gives plenty of room to maneuver: I would be still happy to offset at 10x the ‘list price’ of the offset to conservative accommodate these effects.
More importantly, one should expect these effects to accrue to donations as well. The people I persuade to refrain from animal products via donations will presumably have similar effects in terms of promoting pro-animal ideas or exerting pressure in their social network. Perhaps within the EA community vegans are particularly important, so that a vegan EA is worth much more than a vegan non-EA, perhaps as steering the EA community in a more animal-welfare direction would be very high impact. Yet considerations along these lines seem to recapitulate discussing the merits of animal causes versus other cause areas: direction seems fairly zero sum between cause areas, and directing more EA-energy to animal welfare is only desirable if this cause area is better than whatever area these energies would otherwise be directed towards.
Mutual non-exclusivity and marginal cost
Another reply would be the juxtaposition between eating meat and giving money to animal causes is a false dilemma. Eating animal products does not reduce the amount of money one can donate (indeed, as plant-based diets are often cheaper, going vegan could available donation funds!) 8 So being vegan and giving the money to animal charities seems better.
Yet all of us fall considerably short of being ideal utility maximizing agents. Whether we talk about an explicit ‘selfishness’ or ‘willpower budget’, we can look at the margin of our current moral lapses and look at what aspects of our lives are best to push on. Given the apparent cheapness of turning others vegan and the difficulty for most of us in changing our diet, it seems easier to work on ways of giving a few dollars more rather than changing our diet. Similar reasoning applies to other arduous sacrifices which are not strictly ‘mutually exclusive’ with our donations (e.g. always having cold showers to help the climate).
Non-consequentialist considerations
The foregoing has only considered consequentialist concerns. There are others. One is an implied inconsistency: offsetting donations only work if there are some people willing to ‘take the hit’ and change their diet. Yet if we all opted to pay money in lieu of changing our diet, no diets would actually change. One could draw uncomfortable comparisons to indulgences, or the much maligned pharisees who paid others to pray on their behalf.
Yet these ‘if everyone did this, what would happen?’ worries tend to go wrong. It would be bad if all charity workers went into banking, but that doesn’t mean that none should. Perhaps this should be viewed more along the lines of moral trade, with people who really like animal products trading with those who really like money. If too many try and offset, the ‘price’ of converting people to veganism may rise too (Scott Alexander neatly explains how this could work here). 9
A greater concern is this. We generally do not accept offsets with other sorts of immoral behaviour. ‘Paying for’ murder with a $10 000 donation to AMF (a multiple of the expected cost per life saved) seems morally outrageous, rather than good moral deal on net. In the same way, if animals have moral status, killing them and eating their bodies seems wrong in a way which cannot be washed away with sufficient offsetting donations. 10
The usual utilitarian reply to these sorts of cases is to appeal to the (very negative) second order effects as prohibitive. In reality killing someone does some injury to the social fabric and trust a community has, would be awful PR for the ’cause’, has a high probability of going wrong, among many other costs. Many utilitarians are ‘rule’-utilitarians for these reasons: it may well be the case that members of society pre-commiting to following certain rules may lead to greater overall utility, than a society where agents contravene commonly-held norms because the judge the expected outcome to be greater. 11 These effects diminish the more generally acceptable something is: the true cost for ‘offsetting’ murder may be vastly more than the object-level of a single life, the true cost for offsetting eating meat, which most members of society do anyway, may fall very close to the object-level costs.
Lots of our actions cause animal harm and death, whether through carbon emissions from non-essential activity, roadkill due to transport of goods, small animals in fields which harvested, or crushing insects when we walk around. Perhaps some harm is unavoidable, yet all (vegans included) could make further boycotts or actions to bring us closer to a minimum, and not doing so could be gravely immoral by the lights of various non-consequentialist concerns. 12 Yet acceding to these further demands appears both arduous and counter-productive.
Perhaps it is a matter of balance, and duty falls somewhere in the middle: it is not permissible to murder for net-good consequences, but it is permissible to use energy or other goods which generate animal deaths as a forseeable consequence. Where veganism falls in this spectrum depends on the particular non-consequentialist concerns being offered, how they interact with precisely what is being done to the animals in question.
Summary
Very optimistic estimates for the effectiveness of animal welfare causes is prima facieinconsistent with a great concern for personal dietary change. Given the small donations required to give an equivalent amount of benefit to personally becoming vegan, focusing on this seems an ineffective use of energy. Far easier (for both advocate and advocatee) to give fairly token donations to animal charities, which do a lot more good.
This essay explored means of bringing these commitments into reflective equilibrium: adjusting the estimate for animal charities down, or personal veganism up, or appealing to non-consequentialist concerns that make stopping consumption of animal products morally imperative despite the modest benefits. I favour a significant downward adjustment of the effectiveness of animal charity effectiveness. Besides bringing the figures of vegan offset closer to sanity, one should expect winner’s curse, reference class reversion and other effects to take orders of magnitude from the point estimates (unbiased though they are). If the cost to offset veganism is one hundred times higher than the estimates, then it becomes a much more desirable addition to one’s altruistic behaviour given the non-monetary costs of dietary change. I think this 100 fold adjustment is about right, and if anything might still be slightly optimistic.
There are interesting sequelae for those who favour different resolutions. If one takes a high view of the importance of second order effects of veganism (intra-EA or not), this implies that precise efforts to quantify the first order effects is less important, and the appeal to their size disingenuous. If one thinks one should be vegan because it can be ‘additional’ to our other altruistic actions, or it is mandated by non-consequentialist reasons, a fuller account is needed to decide which circumstances (if any) one can knowingly cause harm, and thus which actions out of all of those available are ruled out.
Notes:
- This is a follow-on from an earlier attempt. The more involved discussion of non-consequentialist approaches I owe to comments on fb and the EA forum in general, and to Carl Shulman’s helpful remarks in particular.
- There is one significant complication: generally these interventions ‘work’ by preventing farmed animals coming into existence in the first place. This may be a bad thing if you are optimistic that farmed animals generally have lives which are on balance worth living, despite their suffering. More exotically, even if you are pessimistic about farmed animals having lives worth living, this may be a good thing if you are even more pessimistic about wild animals having lives worth living, as farmed animals could displace wild animals who suffer even worse lives.
- Hurford offers the following comparators:
Giving up beef is ~2.1x as important as giving up dairy.
Giving up beef is ~1.9x as important as giving up turkey.
Giving up pork is ~2.9x as important as giving up beef.
Giving up chicken meat is ~11.3x as important as giving up beef.
Giving up eggs is ~11.8x as important as giving up beef.
Giving up aquacultured fish is ~6.4x as important as giving up beef.We can use these multiples to put everything in terms of ‘dairy units’, and thus calculate the fraction of harm caused by particular diets:
Dairy = 1 Beef = 2.1 Turkey = 1.1 (2.1/1.9) Pork = 6.1 (2.1*2.9) Chicken = 23.7 (11.3*2.1) Eggs = 24.8 (11.8*2.1) Fish = 13.4 TOTAL = 72.2 Fraction of harm (Full diet = 1, Vegan = 0): Vegetarianism = Dairy+eggs/total = (1 + 24.8)/72.2 = 35.7% Lactovegetarianism = Dairy/total = 1/72.2 = 1.4%
- Another benefit is this will implicitly cover all impacts of reducing animal products, which may be important for those worried about the environmental impacts.
- This isn’t perfectly conservative: data on ‘how long do people stay vegan’ is extrapolated from ‘how long do people who limit animal products continue limiting’, and I’d speculate vegans might have higher rates of attrition than vegetarians. That said, the other adjusts are probably conservative.
- Despite my original post, I’m less interested in moral offsetting as such. On thoroughgoing consequentialism, offsetting makes little sense. Either the ‘offset charity’ is the best way to spend your money or it isn’t. If the former, you should donate to it anyway; if the latter, it would be better to give the offsetting money to the better cause. Perhaps better to use the ‘cost of veganism’ comparisons as a metric for comparing different moral behaviours, without demanding explicit offsetting. If being vegan is much more challenging than earning an extra hundred dollars, working on the latter seems a much better strategy.
- As Carl notes elsewhere, the current vegetarian and vegan population being of a significant size speaks against really big second order effects. Even on an exponential growth model, the present size of the vegan and vegetarian movement speaks against very fast growth rates, and a high expected ‘R0’ for being a vegan.
- But note this consideration can easily flip the other way, especially in the case of milk given how cheap it is to offset – if vegan substitutes for milk cost a tiny amount more, it becomes better to just drink cow milk and donate the cost savings.
- One may worry about other adverse consequences of direct moral offsetting, as it might be a hurdle towards dramatic social change in attitudes towards animals. Yet the sign of this effect is unclear – a wider recognition that eating animals is something that demands moral compensation may be important, and might be spread more easily if ‘buy-in’ is small donations rather than dietary change. Further, if one thinks consequentialising is good, widespread use of these things might be good too.
- Although it is perhaps worth noting non-consequentialist moral theories might assign moral status on non-welfarist grounds (e.g. ability to have interests, plan for the future, having a soul). Animals are probably at greater risk of ‘failing’ these tests for moral status, and consequently killing or eating them may not be morally significant.
- There is more that can be said. An objector could turn the screw by offering caveats to the thought experiment (or just sufficient offset) that the harm of killing is genuinely outweighed by the benefits – yet, they insist, doing so would still be wrong. Although borderline question-begging, an objector might think the utilitarian gloss as to why ‘killing offsets’ are immoral is insipid, and the real reasoning should rely on the intrinsic immorality of killing.
I think consequentialists should ultimately yield – these are costs to the theory, and the best that can be hoped for is to amelliorate these (one could appeal to saying ones intuitions in a sufficiently caveated case are unreliable may be one approach). It would be remarkable if a moral theory on reflection accorded with all of our moral intuitions all the time. Instead of defending the difficultly defensible, it is perhaps better to appeal that the balance of benefits of costs of consequentialism does better than other theories, notwithstanding its poor performance in these particular cases.
- One may say there is something different in these cases: although the deaths and suffering of animals is forseeable by-product in other cases, this is less objectionable than directly ‘commissioning’ the harm by eating meat, which demands the animal be killed. Perhaps, but even if so the former would be a type of wrongdoing, even if of a milder quality – and maybe not significantly milder: knowingly performing acts that lead to someone else’s death is generally tantamount to murder. Further, it is unclear how much this distinction matters – it might license eggs and milk being ‘not as bad’ as the harm to male chicks in egg farming might be similarly indirect, and my hunch is we can find similarly ‘direct’ examples of harm to animals that do not involve diet.
Question: Does anyone know how big the pool of farm-animal charities ACE has considered actually is? And do we know if the best ones are a lot better than the mean?
The reason I'm asking is that the amount of regression to the mean/ winner's curse we should expect depends (mostly) on two things: How big is the sample out of which we've selected the best charities and what's the actual mean effectiveness in the appropriate reference class? (Variance would also be a factor)
To make this more concrete, suppose you let 4 people do a run and pick the fastest. If you let them run again, the previous winner will probably be roughly as fast as the first time. But if you did the same experiment with 1,000,000 people, the first winner would probably be significantly slower the next time because they just had a really good day the first time.
My guess would be that the number of charities we could call part of the sample is lower for farm animal advocacy than for global poverty, which would mean that we should expect less regression to the mean.
The second point is that in case the best charities in the farm animal advocacy sector aren't actually that far above the mean, we shouldn't expect our estimate of their effectiveness to decrease as much as we get better information.
Looking at the ACE website, they recommend 3 and examined 100 organizations. The cost-effectiveness estimates for the top charities seem to be heavily based on the same studies (with related methodological issues) suggesting that veg*n ads/flyers have incredibly high persuasion rates. The analysis of media impressions also uses those same studies to estimate how much meat reduction media coverage will give. There is also some discussion of policy changes and legal activity, as well as RCTs, but they place less of a role in the cost-effectiveness estimates.
So it looks like the reason the estimates are so high is because of the claims that 2% of people reached with ads/flyers will give up meat, combined with selection of organizations whose activities are evaluated in terms of ads, flyers and videos. I would worry about regression on the studies more than on the selection of charities conditional on the studies being right.
A small quibble
I think it's flat out wrong to conclude our diets "are no big deal". Being vegetarian for a lifetime prevents over 1000 years of animal suffering. That's a huge, huge impact.
My more serious worry is that people will draw this conclusion and eat less ethically as a result, without donating more (they already knew donating was great). But this is just psychological speculation backed up by some anecdotal evidence.
Interesting post. I have a minor objection to this sentence, though:
The way I interpret this, you take the fact that these commitments aren't in reflective equilibrium as a (weak) reason to adjust the estimate for animal charities down (you claim that the inconsistency forces us to either adjust the estimate, or one of the other commitments). But as I see it, the fact that those estimates aren't aligned with your moral views is not even a weak reason to adjust them down. Beliefs about facts, such as these estimates, should be based on evidence (which you do provide in the rest of the post, by all means), not on whether they are aligned with our moral views or not.
Beliefs about facts aren't just based on evidence, because you have priors before you have evidence. If I have a prior belief that converting people to veg*nism is valuable and the ACE estimates then contradict my belief by making it so cheap to buy that converting is nearly pointless (or at least, pointless without appeal to some other upsides that I happen to find unconvincing), that is at least a weak reason to be sceptical of the ACE estimates.
I think that summary describes most of the people Greg is addressing here adequately.
Suppose I have a prior belief that it is valuable to convert people to Christianity. I then get an evidence-backed estimate saying that the Christian teachings are probably false. That contradicts my belief that I should try to convert people to Christianity. Should that count as a reason to reject that estimate? Obviously, no. To reject the estimate you need evidence, not merely stating that it is incompatible with your moral views. The same holds for the veganism question.
"Beliefs about facts, such as these estimates, should be based on evidence"
"Beliefs about facts aren't just based on evidence, because you have priors before you have evidence."
"To reject the estimate you need evidence"
I'm not sure from this sequence what in my comment you actually disagree with; you appear to have simply chosen a different analogy and then restated your position. If the evidence comes in way against your prior, do you disagree that that is on its own a reason to be sceptical?
I also think you're drawing a false dichotomy between 'moral views' and 'facts'. Believing animals matter is generally considered a moral view, but many animal welfare advocates believe it almost directly follows from factual observations. Believing it's valuable to convert people to x is mostly a factual view, but tends to be grounded in moral views around what makes x good in the first place.
http://lesswrong.com/lw/ig/i_defy_the_data/
Someone once presented me with a new study on the effects of intercessory prayer (that is, people praying for patients who are not told about the prayer), which showed 50% of the prayed-for patients achieving success at in-vitro fertilization, versus 25% of the control group. I liked this claim. It had a nice large effect size. Claims of blatant impossible effects are much more pleasant to deal with than claims of small impossible effects that are "statistically significant".
So I cheerfully said: "I defy the data."
...
Oh, and the prayer study? Soon enough we heard that it had been retracted and was probably fraudulent. But I didn't say fraud. I didn't speculate on how the results might have been obtained. That would have been dismissive. I just stuck my neck out, and nakedly, boldly, without excuses, defied the data.
Yes, the analogy is a re-statement of my position, using another example. I hoped that by using the Christian example, it would become clear that it's not right to reject evidence because one is morally uncomfortable with it. In my view, that amounts to political bias or wishful thinking.
I definitely do believe that there is a dichotomy betwen moral views and facts, and think that is an integral part of the scientific world-view. But leaving that huge philosophical issue aside, I think that if Effective altruists accept these kinds of arguments, we will be far less effective. Faced with our cost-effective estimates of, e.g. Against Malaria Foundation vs ALS, non-EAs could always say that those estimates contradict their preference to give to ALS, or whatever charity they feel like giving to, and that they therefore choose to reject the EA cost-effectiveness estimates.
Another pragmatic argument for why we should strive to be as objective as possible is that unbiased, objective science, seems to have been very effective at acquiring new knowledge, whereas politicized, biased science has not, as Noah Smith points out.
This is actually an important question. In my view, the notion that you can't reject facts that you feel uncomfortable with for moral or political reasons is an important tenet of Effective altruism.
You're mixing several claims here. One is about moral views versus facts, which I agree is a large philosophical discussion. Let's put that to one side.
Another claim is about the proper role of priors. If I'm reading you correctly you think there is no role for priors in evaluating claims whatsoever. That's a surprising position, to put it mildly, so I want to make sure I am reading you correctly before engaging with it. I still don't know whether I am.
Then there's a number of claims about how bad political bias is and why we should try to avoid it, which I agree with and consider at best tangential to the discussion at hand. Striving for objectivity is not the same as ignoring common sense.
No, I'm not ignoring priors. If I have a strong prior belief that my dice is unbiased, clearly I shouldn't give that up because I roll three sixes in a row (although you should adjust it slightly downwards). Factual priors should influence your factual beliefs.
What I am saying is that you shouldn't let your moral views influence your factual beliefs, and that doing so amounts to bias. Hence the whole bias/objectivity issue is very relevant here.
I have a concern that is not directly addressing this article but is related to. I think it the amount of animal-based product you consume also matter. As an ovolactovegetarian, I realized that the amount of dairy products that I consume has increased in comparison to when I was eating meat. Eventually eating just a bit of high-quality meat would lead to far less consumption of dairy products and is better off. It would be interesting to have some quantitative statement about this.
Is this a duplicate of "Don't Sweat Diet"?
Greg mentions that it's a version adjusted in light of feedback. :) For instance, it discuss the possibility that the cost-effectiveness estimates of veg*n outreach could be large overestimates, as raised by you and me.