An integrated model to evaluate the impact of animal products

Edit: this is out of date, see https://​​fo­rum.effec­tivealtru­​​posts/​​YuFD4v7DFBcM57eSA/​​con­se­quences-of-an­i­mal-product-con­sump­tion-com­bined-model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://​​​​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.


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://​​​​x/​​s!At2KcPiXB5rkuxZxET3jFz6Jrtr5