Edit: this is out of date, see https://forum.effectivealtruism.org/posts/YuFD4v7DFBcM57eSA/consequences-of-animal-product-consumption-combined-model
This Excel model combines regular and cross-price elasticity effects, welfare analysis, near- and long-term pollution to evaluate the relative harms (or benefits) of different animal products. You can look at it, or save a copy to modify it however you wish. It has some columns that you can use to evaluate a specific diet.
Link to the spreadsheet: https://1drv.ms/x/s!At2KcPiXB5rkuxZxET3jFz6Jrtr5
Sources for the more straightforward inputs are given in the spreadsheet. In this post I will describe/justify some of the more subjective inputs and elements of the model structure.
Elasticity effects regularly refer to the shift in quantity supplied when one consumer's demand increases. Other consumers buy less because of the higher price, partially offsetting one's decision. The cross-price elasticity refers to the increase in other products which are purchased as substitutes by these other consumers. If I buy 1kg of chicken, some people will buy less chicken, and some of those people will buy more of other things (e.g. turkey).
Unlike the regular effect, cross-price effects haven't been properly calculated (as far as I know). I basically guessed them for each animal product based on base rates of animal product consumption, the typical relative prices, and dietary patterns. These estimates are really not robust but they are better than nothing, feel free to replace with your own.
I found no satisfactory scale of moral weightings. Probability-of-sentience estimates given by Luke Muelhauser are inadequate because they assume that all sentient animals have equal sentience; the limited cognition and perception of simpler creatures should be considered even if we assume that they are sentient. Insufficient data is available for neuron count, and brain mass seems clearly wrong (it leads to elephants being much more important than humans, and cows much more important than pigs). Instead, I made my own estimates, based on reading basic information about the behavior and cognition of different animals, using my subjective-well-being perspective.
For quality of life I did not use Brian Tomasik's calculations, because they only quantified suffering and ignored happiness. I created a second sheet to integrate multiple groups of estimates of animal quality of life; the first is the ratings by Charity Entrepreneurship and the second is the ratings by Bailey Norwood in his book Compassion, by the Pound. I gave more weight to the former because it has multiple people's inputs, a more rigorous formal system and is done by EA researchers. However, there is a systematic difference between the two, as Norwood was more optimistic about animals' quality of life. There is also more optimism about farm animal lives coming from farmers, who are more familiar with them than anyone else. Therefore I kept the weighting for Norwood's estimates almost as high, in order to adjust for this bias that we get when we only hear one side of the story.
Neither group of estimates had estimates for every type of animal, so I created dummy estimates. First I measured the average difference in quality of life evaluations for animals which were evaluated by both sources (53, on a +100 to -100 scale), then I added or subtracted it from one source to fill in the blanks of the other. Because this is less robust than the real evaluations, I gave the dummy estimates less weight.
Both of the scales seem to have a flaw in that they are symmetric from suffering to happiness. Even under a standard utilitarian view, people and animals seem to have more capacity for suffering than for happiness; a perfect life is not enough to outweigh a terrible life. The ideal way to address this would be to redo the evaluations from the ground up, but for my purposes I used a rough workaround of adding an input parameter which lets you overweight animals with net average suffering while underweighting animals with net average happiness. This is not really precise because it is an ex post facto modification which neglects the variance in welfare within and among animal lives, but it seems to be a decent approximation (especially since the animals are mostly living in net suffering anyway, per these estimates). The default value (2) is my guesstimate from a standard utilitarian framework; more substantially 'negative-leaning' views can be modeled by providing a higher number.
The short run costs of greenhouse gases turned out to be very small and straightforward to compare with animal suffering, but the long run impact to our economy and society are a different story. If climate change hurts our economy then we may be shifted on a fundamentally slower trajectory of civilizational progress. As far as I know, there isn't a satisfactory estimate of long-run societal utility, so I attempted my own using the Doomsday Argument. Assuming a noninformative prior distribution over the eventual number of humans, the expected number of future humans is equal to the number of past humans (108.5 billion). Using the standard Self-Sampling Assumption where the observer is selected from all observers, we must input the expected lifespan of future people to calculate future utility. I chose 120 for a future with a small minority of (theoretical) immortals and/or widespread modest improvements in longevity.
If you use the Strong Self-Sampling Assumption, the observer-moment is selected from all observer-moments. You can model this with the spreadsheet - just use the average historical lifespan for the expected lifespan of future people.
We may have significant doubts about the Doomsday Argument that lead us to defer to a more basic direct estimate of expected future population and lifespans. First, the DA might be philosophically wrong, and second, humans might evolve or be replaced by agents that fall outside our reference class. In these cases, you can estimate a larger future population and/or longer future lifespans.
Conclusions
After viewing my results and also testing out plausible variations in the inputs, these stuck out as the main conclusions.
- Milk is essentially unobjectionable. Even after estimating cross-price elasticity effects and environmental damage, the impact of milk is comparatively negligible. In fact the well-being of the consumer, not included in this model, may outweigh the other effects of the product. (Veal production is so low that marginal milk production should do nothing to change it; Compassion, by the Pound points out the massive fall in veal production over time.) The EA community should not push veganism except insofar as a milk exception is considered weird and difficult to communicate.
- The beef+milk+plants diet that is often suggested as an easy way to reduce animal suffering should be dropped. Even under fairly conservative assumptions about long run climate change and social utility, the costs of beef are higher than some other animal products. A regular vegetarian diet appears to be generally better, and there is substantial uncertainty over whether beef or chicken is better. A more plausible set of inputs (the defaults in this spreadsheet) ranks beef near the bottom, and things get astronomically worse if we drop the Doomsday Argument in favor of a simpler direct estimate of future social utility.
- Giving up fish is still extremely important under regular assumptions, arguably more important than the much harder step of moving on to a vegetarian or vegan diet. We should consider ways of targeting our activism more specifically towards fish consumption and production. This has been said before, but it remains robust under this model and needs to be stressed.
- Short term animal suffering is the main issue under most outlooks, and dominates the issue if you have a serious expectation for some impending extinction event. Climate change concerns dominate short term welfare once you include trillions of people or millennia-long lifespans in your framework.
- Short term morbidity and mortality from climate change seems to be much less significant than the long run slowdown in economic growth and expansion.
Link to the spreadsheet: https://1drv.ms/x/s!At2KcPiXB5rkuxZxET3jFz6Jrtr5
Thanks for putting this together! It's really great to see attempts at quantitatively prioritizing animal suffering, and I wish people would do more of it.
I do think you are misusing the word elasticities. The table you cite in compassion by the pound used the price elasticities of supply and demand to come up with their estimates, but these numbers they report are not elasticities, since that term usually refers to the change in demand or supply in response to price or income.
This leaves me particularly confused as to your interpretation of what you call "cross-price elasticities"--a cross-price elasticity of demand refers to the percent increase in demand of commodity x in response to a percent increase in price of commodity y. You are using the term to refer to "the increase in other products which are purchased as substitutes by these other consumers." The reason this distinction is important is that traditional elasticities of demand account for individual consumers substituting one good for another due to price changes--they do not show the net change in equilibrium quantity of one good in response to a demand change of another good. Since you are basing this on your own assumptions rather than data, all I'm suggesting here is a rephrasing.
If you are truly looking for a cross-price elasticity of demand, these are well studied for most commodities and can be found here.
My final comment is that I am puzzled by your conclusion regarding milk given that the welfare metrics you use are just scales from better to worse and do not have an interpretation for their absolute value. It could be that your metrics are exactly spot on, but milk products still impose enormous suffering on cows per pound. I'm puzzled by your conclusions regarding the relative importance of climate change for the same reason.
Anyways, thanks again for working on this. I hope my comments don't come across as too critical, as I think that carefully reasoning through these issues is really important.
Ah, yeah, thanks for underlining that, I say "elasticity effect" as opposed to elasticity, maybe that's not clear enough.
The quality of life evaluations both have 0 for a life with neutral value, if that's what you mean.
Each point in the scoring corresponds to 1/100 of the difference between a neutral life and a life with all interests satisfied, for a human, for one day.
So the estimated net harm of a cup of milk is like feeling 1% worse for a day, on my assumptions.