Direct suffering Guesstimate.

Brian Tomasik’s classic “How Much Direct Suffering Is Caused by Various Animal Foods?” investigates the most easily observable suffering footprint of various products derived from animal products.

The page, in it’s latest incarnation, contains a JavaScript calculator for putting in one’s own values. But I’ve gotten used to using ranges and having a sensitivity analysis to draw on to decide which cells to research further. So I transferred Brian’s model to Guesstimate, mostly drawing on his research but also adding some of my own when I couldn’t decide what the variance should look like.

Finally, I’ve also added some more data points to it:

  1. I researched the average consumption (in the US). People tend to think in portions rather than in kg, and they tend to think in terms of cutting out animal products rather than reducing their consumption by an absolute delta. The new suffering times average consumption column greatly reduces the gap between beef and milk because people tend to consume so much milk, and it makes chicken meat worse than eggs because people tend to consume so much chicken meat.

  2. I’ve added a row for organic eggs. I’ve often been asked how organic eggs stack up against battery cage eggs (or even more ambiguously, how aviary system eggs stack up). I never knew, because while the conditions are arguably better,1 the chickens used are usually not Leghorns, so they produce fewer eggs per year and at more steeply declining rates so that they’re killed earlier. Therefore, more chickens are needed to satisfy the same demand.23

    I still don’t know – my estimate is much to close and unreliable for that – but it looks like organic eggs fare a bit better. The difference is less than even just one order of magnitude at the moment, so expect it to change when you touch some of the more sensitive inputs.

Brian’s Reservations

In 2014, Brian added some reservations to his article:

I’m somewhat less gung-ho about these numbers than when I first wrote this piece because in practice, the side effects of meat consumption on wild animals and Earth’s long-run future probably matter much more than the (horrific) direct impacts on farm animals themselves. Of course, evaluating the net impact of these indirect side effects is much trickier.

Whatever the sign is of the indirect effects, indirect effects should be more similar across animals than the suffering figures are across animals, since cows and chickens don’t differ as much in their environmental impacts as in their direct suffering per kilogram. Hence, these neglected factors should tend to drive the (absolute value of the) ratios of per-kilogram impact estimates across species closer toward 1.

Read The Importance of Wild-Animal Suffering for more information.

Faunalytics Research

Faunalytics has done a very similar investigation. They probably did more detailed research and used different and finer categories, which seem to match more closely categories of animal product–based dishes than the animal products that go into them. They present point estimates rather than ranges and opted not to weigh the intensity of the suffering:

As one final note on the methodology, some readers may have encountered other impact estimates where subjective multipliers are applied to the amount of suffering experienced. We did not do so, treating each day of life as one day regardless of the quality of that life. Although we believe that differences in quality of life and suffering are probable, biases due to anthropomorphization or lack of sufficient data are also likely and, in our view, more problematic. It is worth noting that many of those animals for whom quality of life is likely lowest (e.g., layer hens, farmed fish) are already high on the impact list for other reasons. Any reader who would prefer to recalculate the estimates with additional subjective days of suffering can do so using the data and code files available on the Open Science Framework.

Assorted Considerations

Most of the considerations are not mine but Brian’s and ACE’s (Dairy/Eggs AYLA and AEPY , Meat Land Animal Equivalents Per Person Per Year, Leafleting Impact Calculator6), so check out their work for more guidance. (I’ve also usually linked and copied their summaries into the Guesstimate cells.)

Leghorn Lifespans

I’ve been reading various different figures on the average lifespans of Leghorns. These chickens can of course live many years, but they produce most eggs in their youth, their first year, and then fewer and fewer with every further year. Therefore, their average age at death is determined by a calculation that probably involves the monetary value of their meat, rent, and the price of new pullets.4 The age that makes for the most profitable killing may be different between countries.

  1. Brian’s Canadian source (before 2007 and perhaps before 2000 judging by the dates of the sources) indicates that they live for around 500 days and lay 288 eggs p.h. p.a. (per hen per year).
  2. Norwood and Lusk (2011) write, “Cage eggs are assumed to be white eggs from a White Leghorn type breed. This hen will produce 509 eggs throughout its 2.21 years of life.” So around 800 days at 230 eggs p.h. p.a.
  3. ACE’s analysis of the USDA statistics from 2012 and 2014 indicates that they live around 412 to 515 days.
  4. My different approach to almost the same data (I only have access to the 2012 statistics) puts the average age at 602 to 634 days (at 271–274 eggs p.h. p.a.).

I discount Brian’s source only because it’s Canadian and over a decade old, so the economic conditions were surely different enough to explain the difference. Norwood and Lusk have their own university farms where they may allow their chickens to get older than they usually do; their figure is the highest of all. I rely most on the USDA statistics and there on my reanalysis.5

Cow Lifespans

Some data points:

  1. Bovine enthusiast: 5–6 years
  2. University of Illinois: “The typical cow remains in the milking herd less than 4 years even though peak milk production related to maturity ordinarily does not decline until 8 or 9 years of age.” Cited by Wikipedia as evidence for a lifespan of four years, but the author probably intended 4 + (1 to 2) years because the cow as to grow up first before she can produce milk. (Does someone want to correct it?)
  3. Brian: Alberta Milk (n.d.): “The typical dairy cow lives an average of five years, with the first two years focused on providing a strong foundation for the healthy development of the cow. From age two, the mature cow will become a productive member of the milking herd (meaning, she will produce milk).”
  4. Penn State indicates that conception happens at around 15 months
  5. Virginia Tech implies that calves are born for the first time at 24 months.

Most confusion seems to stem from the time the cow has to grow up before she can be artificially impregnated by the farmer, and then the time that her child grows up in utero before lactation starts. But there are still differences of three to four years even so. The result is not very sensitive to this input, so I haven’t researched it further, but if someone know what the most reliable data are, then I’d love to adapt the Guesstimate!

Sentience Multipliers

For the sentience multipliers – numbers from the interval [0.0, 1.0] – I opted for something that captured my conflicting intuitions about sentience and looked properly arbitrary at the same time so not to signal sophistication where there is none. The formulae ended up looking like =max(0, 1-lognormal(2.5, 1)/100). A log-normal distribution mirrored at a vertical axis so that it bulges up right before 1, and crudely tweaked so not to go below 0.

My intuitions are (1) that sentience may come in degrees, may reach different degrees for different individuals of the same species, may fluctuate for the same individual; (2) that experts disagree over the sentience of different species so that they have degrees of sentience with different probabilities; or (3) that people can legitimately have different opinions on the degree of sentience of an individual just as they can have differing opinions on how democratic each of the US, Switzerland, China, and North Korea are.

In each of these cases, my distributions mean different things, but the general shape seems to match all of them similarly well or badly.

  1. Chickens farmed for meat are killed very young, after around six weeks, so they receive a slightly lower sentience distribution than chickens farmed for eggs, because I think I was less sentient as a child too.
  2. Fish like to cause little scientific kerfuffles (just search ACE’s Research Library for “fish”) over whether they’re conscious or feel pain. Someone who has read this far is probably acquainted with expected value and will not use the possibility that fish might not suffer as an excuse to buy and eat them. That said, I was surprised by how sensitive the result was to this input. It still takes a probability of 0.0001 (or odds of 1:10,000) to press the expected suffering per kg for salmon to the level of milk, and there are probably nowhere nearly enough fish experts in the world to be sure at such an extreme level even if they all agreed, but as it is, the input is correlated with the result at r² = 0.58 (and looks correlated too in the scatter plot).

Suffering Multipliers

The suffering multipliers were somewhat informed by Dr. Sara Shields’ and Dr. Bailey Norwood’s estimates cited in Veganomics (and, for the latter, in Compassion, by the Pound).7

  1. For the most part they agree nicely, but there’s the controversy over whether chickens farmed for meat have lives worth living. On this issue, they have strongly different opinions. My intuitions are informed by theirs, but seeing that a flu has a disability weight > 1 for me, you may make sure that you agree here. In particular, if you’re a classic utilitarian (of the rat god variety ;-)), this may make a sign-changing difference for you.
  2. People are also notoriously unsure whether or to what degree Dr. Norwood factored in such things as sentience, transport, slaughter, slaughter-to-lifespan ratio, etc. in his estimates – that would be double-counting in our case. I don’t know anything about the context of Dr. Shields’ estimates, so I’ve tried not to rely on them too much.
  3. The model only covers suffering, so the probably positive lives of cows farmed for meat is not captured. If offsetting happiness is a thing for you – as it may be for me in this case – then the result for beef is probably uninformative for you.
  4. Note also that I’ve created a bit of arbitrary fuzziness around the value of 1 for the cow farmed for meat because different individuals are bound to have different life experiences.


  1. For example, according to FOWEL – and yes, I’m aware of the big debate around this recently, but I’m from Europe where no one believes that there’s something worse than battery cages. So I’m shifting the burden of proof to DxE here. ;-) (In reality, I don’t think I have any influence on the prioritization decisions of Open Phil, Mercy for Animals, The Humane League, et al., so I tuned out of the debate to focus on things that I can affect.)

    Norwood and Lusk (2011) on the topic: “In the FOWEL model, the cage system receives a score of zero [out of ten], a barn system (barn with hens uncaged on the floor) receives a score of 5.9, and an organic system scores 7.8.” They continue: “Much of the differences in opinion can be explained by different perceptions of the importance and magnitude of the mortality rates. Based on the available evidence, our estimate of the mortality rate in cage systems is about 3 percent. By contrast, we estimate mortality rates of 7 percent in cage free systems and 9 percent in free range systems …. Organic systems have higher mortality rates of about 13 percent because of feed restrictions. Organic producers cannot supplement animal feed with ‘unnatural’ synthetic (man made) amino acids. Another obstacle is the fact that a farmer cannot treat a sick animal with antibiotics and then sell the animal for organic food. This causes some farmers to deny antibiotics to sick animals. As a result, hens suffer. A number of animal scientists in the US believe organic production is cruel to hens for this reason.” 

  2. I’m completely ignoring supply and demand elasticity here, because a shift to organic production would lead to a slight price increase due to the production and a big price increase due to warm-fuzzy premiums for retailers, thus probably reducing consumption. I’m ignoring these, because my audience are individuals potentially much more motivated by other factors than price. (Also, the calculations would get a heckuvalot more complex.) 

  3. This tradeoff makes sense from a classic utilitarian or negative utilitarian perspective, but it doesn’t so much for me. I perceive some sort of threshold between milder discomfort and extreme suffering, and think about milder discomfort more in a classic utilitarian way (where it can be traded off against happiness) while I think about extreme suffering more in a prioritarian way (where it is lexically worse than any happiness could be good). Maybe it’s not lexical but just a really big jump, not sure. So if in one system few chickens suffered extremely (above the threshold) it would be lexically worse than one where many chickens suffered mildly (below the threshold), but neither would necessarily be of positive net value. But my threshold to extreme suffering is low (even a flu has a disability weight > 1 for me), so it probably doesn’t make a difference in this case. 

  4. One advantage of having vegan friends is that I don’t have to worry that they might eat me. Another, it seems, is that I won’t have to worry that they might kill me off to conserve government funds when I reach retirement age. 

  5. ACE’s result seemed really surprisingly low to me, so I figured there’s a chance that ACE might’ve gone wrong here. I’m still not sure, though.

    I only have access to the 2012 USDA statistics (which cover 2011) and created a spreadsheet for the calculations (disregard the second sheet for now).

    ACE’s approach was to divide the monthly totals of all layer hens “on hand” (stated prominently at the beginning of the report) by the monthly rate of “disposal” of all sorts – selling to slaughter, destruction of corpses, etc.

    But on page 46, the report seems to distinguish “layers,” “pullets,” and “other chickens,” implying that whenever they talk about “layers,” they’re not referring to all chickens but only to the ones at sufficient age to lay eggs and also excluding whatever “other chickens” is. But the rate of “disposal” of chickens may (and I’m genuinely unsure here) not make that distinction, so that the total we should be using is not the total only of layers but that of layers, pullets, and other chickens all added up. This total is about 10 million higher in the mean over the two years.

    Using the rate of replenishment and the (higher for some reason) rate of “disposal” to estimate the months of age of a chicken when it is killed, I get the two-year averages of 19.78 months and 20.83 months respectively. With months of around 30.4 days, that’s 602–634 days or 1.65–1.74 years. I’ve used these higher numbers in the Guesstimate for now. But note that this input is just about uncorrelated with all outputs, so it doesn’t really matter either way. 

  6. Sorry for using the dirty word, but it’s about much more than just leafleting! Please don’t dox me or call me names because I’ve said “leafleting.” Aaah! 

  7. Disclaimer: I don’t condone the author’s conduct. (I have no first-hand information, but the description of a former employee matches gender-based gaslighting.) Cooney authored several informative books, so that I can’t help but cite him occasionally lest it seem that I plagiarize them. But I’m adding these disclaimers to avoid the impression that I accept such conduct or that it is accepted in my circles. 


comments powered by Disqus