Causal Network Model III: Findings

This is a writeup of the find­ings from the Causal Net­works Model, cre­ated by CEA sum­mer re­search fel­lows Alex Barry and Denise Melchin. Owen Cot­ton-Bar­ratt pro­vided the origi­nal idea, which was fur­ther de­vel­oped by Max Dal­ton. Both, along with Ste­fan Schu­bert, pro­vided com­ments and feed­back through­out the pro­cess.

This is of a mul­ti­part se­ries of posts ex­plain­ing what the model is, how it works and our find­ings. We recom­mend you read the ‘In­tro­duc­tion & user guide’ post first be­fore this post to give the cor­rect back­ground to our model. The struc­ture of the se­ries is as fol­lows:

  1. In­tro­duc­tion & user guide (Recom­mended be­fore read­ing this post)

  2. Tech­ni­cal guide (op­tional read­ing, a de­scrip­tion of the tech­ni­cal de­tails of how the model works)

  3. Find­ings (this post)

  4. Cli­mate catas­tro­phe (one par­tic­u­larly ma­jor find­ing)

The struc­ture of this post is as fol­lows:

  1. Sum­mary of im­por­tant findings

  2. High­lighted ar­eas for fur­ther research

  3. Effects of spe­cific inputs

Some of the re­sults listed in the last sec­tion are fairly minor, so read­ers may wish to fo­cus only on the first two sec­tions.

The model is both very sim­plified, and many of the re­sults de­pend to a large ex­tent on par­tic­u­lar vari­ables with val­ues about which we have very lit­tle in­for­ma­tion. Be­cause of this, and be­cause of the gen­eral limi­ta­tions of the model, any find­ings should be taken as at most in­vi­ta­tions to fur­ther re­search, rather than as con­crete pro­nounce­ments of effec­tive­ness. (In the past we have found a fair num­ber of mis­takes in­volv­ing num­bers be­ing wrong by a few or­ders of mag­ni­tude!)

For the sake of space and read­abil­ity we largely omit these qual­ifiers through­out the rest of the post, al­though will bring it up when re­sults seem par­tic­u­larly un­cer­tain.

We strongly recom­mend you read part I (par­tic­u­larly sec­tions 2 and 5) to get ap­pro­pri­ate back­ground for the model be­fore this post.

1. Sum­mary of im­por­tant findings

1.1 Steak­less Salvation

Even with very small prob­a­bil­ities of suc­cess, re­search into de­vel­op­ing cost-effec­tive farmed meat al­ter­na­tives (‘clean meat’) can be cost-com­pet­i­tive com­pared with other an­i­mal welfare al­ter­na­tives.

This is due to the po­ten­tial for ‘clean meat’ to gain a large pro­por­tion of the mar­ket share very quickly once it is lower in price but equal in qual­ity to con­ven­tional meat. This means that it will scale far bet­ter than most an­i­mal in­ter­ven­tions. Ad­di­tion­ally this seems to be po­ten­tially a very effec­tive way to re­duce cli­mate change, which has many other benefi­cial effects, as ex­plained be­low.

This is true even for rel­a­tively limited forms of ‘clean meat’, as in our model we only con­sider the pos­si­bil­ity of de­vel­op­ing cost-com­pet­i­tive ‘clean’ ground meat, which seems much more at­tain­able in the short term; this is still effec­tive enough to seem plau­si­bly bet­ter than con­ven­tional an­i­mal out­reach.

This all as­sumes that the mar­ket is not ac­tively hos­tile to clean meat, and all else equal will sim­ply chose the cheaper op­tion. This seems par­tic­u­larly ap­pli­ca­ble in the ground meat case.

1.2 Cli­mate Catas­tro­phe:

Cli­mate change seems to be a much big­ger prob­lem than most peo­ple nor­mally con­sider, both due to the po­ten­tial dam­age caused by the Earth’s tem­per­a­ture ris­ing 2-3 de­grees, as well as the tail risk of run­away warm­ing be­ing a global catas­trophic risk.

Un­for­tu­nately, many typ­i­cal EA ac­tivi­ties (e.g. giv­ing more re­sources to the global poor, im­prov­ing farmed an­i­mal con­di­tions) prob­a­bly cause in­creases in CO2 emis­sions, and so could po­ten­tially be nega­tive over­all due to the effects of cli­mate change. This ar­gu­ment is par­tic­u­larly wor­ry­ing if you think the po­ten­tial of the far fu­ture morally dom­i­nates de­ci­sion-mak­ing. For more dis­cus­sion and elab­o­ra­tion on the cli­mate x-risk con­nec­tion see Part IV.

1.3 Cagefree Costs

One ex­am­ple of how con­sid­er­ing cli­mate change could cause an ap­par­ently pos­i­tive in­ter­ven­tion to be nega­tive is cor­po­rate out­reach fo­cused on an­i­mal welfare. In par­tic­u­lar this af­fects Mercy For An­i­mals’s suc­cess in 2016 in get­ting large busi­nesses to pledge to change from bat­tery to cage-free eggs, af­fect­ing a to­tal of 80 mil­lion lay­ing hens a year. While this is a large win for an­i­mal welfare, cage-free hens are some­what less effi­cient, caus­ing more CO2e emis­sions per egg. This effect is large enough that for ev­ery year ear­lier MFA caused this to hap­pen com­pared to when it would have hap­pened oth­er­wise, it will cost (very, very ap­prox­i­mately) 500 QALYs due to death and dis­ease from cli­mate change be­fore 2050.

This means that if you value a chicken-QALY at less than 120,000 of a hu­man QALY, our model out­puts this in­ter­ven­tion as neu­tral, or even nega­tive. [1]

1.4 Ex­is­ten­tial Effectiveness

Us­ing our de­fault es­ti­mates of the chance of ex­is­ten­tial risk and re­searchers’ abil­ity to re­duce it (or even es­ti­mates or­ders of mag­ni­tude lower), ex­is­ten­tial and global catas­trophic risk re­search and policy work dom­i­nates other cat­e­gories in terms of value. This is true even when com­par­ing to other in­ter­ven­tions in terms of QALYs saved be­fore 2050. [2]

There­fore, you could jus­tify giv­ing to ex­is­ten­tial risk re­search char­i­ties even if you took a per­son-af­fect­ing view or strongly dis­counted fu­ture lives, as long as you put enough chance on re­search be­ing able to re­duce the risks. (That said, if you value guaran­teed im­pact over high ex­pected im­pact, then e.g. global poverty char­i­ties might still be more at­trac­tive).

1.5 Mo­ral­ity Matters

Many of the ac­tions con­sid­ered in the model end up be­ing pos­i­tive un­der some moral the­o­ries and nega­tive un­der oth­ers. Ex­am­ples in­clude the farmed an­i­mal welfare case set out above, or ac­tions that could nega­tively af­fect the far fu­ture while pro­vid­ing value to­day. This sug­gests that char­ity recom­men­da­tions should be more de­pen­dant on the par­tic­u­lars of peo­ple’s moral the­o­ries.

De­spite seem­ing ob­vi­ous when stated, this seems to be some­what at odds with how the EA com­mu­nity ac­tu­ally op­er­ates, where char­ity eval­u­a­tors etc. don’t re­ally talk much about moral­ity when giv­ing recom­men­da­tions.

1.6 Larder Logic

Whether one ex­pects in­ter­ven­tions that re­duce the num­ber of farmed an­i­mals to be pos­i­tive or nega­tive of­ten de­pends on whether one thinks fac­tory farmed cows have lives worth liv­ing. This is also true for other an­i­mals, but cows seem to the biggest case for dis­agree­ment.

2. High­lighted ar­eas for fu­ture research

As dis­cussed in the first post, many of the model’s re­sults de­pend to a large ex­tent on val­ues we know very lit­tle about, and there are many im­por­tant ar­eas of the world we were not able to in­clude in the model due to com­plex­ity or time con­straints.

In par­tic­u­lar, the model sug­gests that it would be very use­ful to learn more about the fol­low­ing ar­eas:

  • How much can re­search ac­tu­ally re­duce the chance of ex­is­ten­tial or global catas­trophic risk?

  • What is the base rate of global catas­trophic or ex­is­ten­tial risks?

  • What is the prob­a­bil­ity of clean meat be­com­ing cheaper than nor­mal meat in the near fu­ture (even just in limited forms, such as ground meat), and how much could ad­di­tional fund­ing in­crease this chance?

  • How well does veg*n out­reach work?

  • Do fac­tory farmed an­i­mals have net pos­i­tive or nega­tive lives? (This seems par­tic­u­larly un­clear for cows.)

  • How nega­tive are the effects of cli­mate change likely to be, and what is the chance of ex­treme run­away global warm­ing act­ing as an se­vere global catas­trophic or ex­is­ten­tial risk?

  • How likely is it that EA will be able to in­fluence gov­ern­ment policy, and how strongly can policy af­fect the things we care about (e.g. x-risk, farmed an­i­mals)?

Some things that we were un­able to cap­ture in the model that also seem im­por­tant to in­ves­ti­gate in­clude:

  • The in­ter­ac­tion of out­reach with (po­ten­tially) time-de­pen­dant ar­eas, and ac­cu­rately mod­el­ling move­ment growth more gen­er­ally.

  • EA’s im­pact on gov­ern­ment in­sti­tu­tions and the effect this has or could have.

  • The value of cause pri­ori­ti­sa­tion (in par­tic­u­lar the likely dis­tri­bu­tions of cause effec­tive­ness and how likely re­search is to find high-value causes).

  • Suffer­ing and the en­vi­ron­men­tal im­pact of fish­ing.

  • Wild an­i­mal suffer­ing.

  • The long-term effects of eco­nomic growth on global in­sti­tu­tions and tech­nolog­i­cal progress.

  • The long-term effects of value changes.

  • Long-term effects more generally

3. Effects of spe­cific inputs

In this sec­tion we go into de­tail about the effects of chang­ing the differ­ent fund­ing in­puts on the user tool, how these effects de­pend on the ‘im­por­tant vari­ables’ and the im­pact of differ­ent moral in­ter­pre­ta­tions of the re­sults.

(In this sec­tion ‘good for cli­mate change’ means re­duc­ing CO2 emis­sions etc.)

Veg*n outreach

  • Low­ers farmed an­i­mal pop­u­la­tions, which is in turn some­what good for cli­mate change and so has a minor pos­i­tive im­pact on hu­mans (our de­faults give $375 per hu­man QALY).

  • Can be nega­tive over­all if you think cows have pos­i­tive lives and are sig­nifi­cantly sen­tient.

  • Depends very strongly on the effec­tive­ness of this veg*n out­reach.

Govern­ment /​ cor­po­rate an­i­mal welfare reform

  • Ex­ceed­ingly pos­i­tive for an­i­mals in terms of in­creas­ing their welfare, re­gard­less of how good you think their cur­rent lives are.

  • Strongly nega­tive for hu­mans (minus one QALY per $100 by de­fault as­sump­tions!) via sig­nifi­cantly in­creased cli­mate change, due to de­creased effi­ciency of cage-free eggs etc. How bad this is strongly de­pends on how bad you think cli­mate change will be (in par­tic­u­lar its chance of be­ing a global catas­trophic or x-risk, and how much it will lower av­er­age qual­ity of life).

  • Still ends up pos­i­tive over­all for views that value an­i­mals >~1/​10,000 of a hu­man.

  • Not very sen­si­tive to any vari­ables be­yond value of an­i­mals vs hu­mans, and those re­lat­ing to cli­mate change.

  • Ad­di­tional un­cer­tainty of how much these in­ter­ven­tions ac­tu­ally help an­i­mals.

An­i­mal Product Alternatives

  • If you think fund­ing an­i­mal product al­ter­na­tives re­search has any real chance of speed­ing up or in­creas­ing the chance of cost-com­pet­i­tive cul­tured ground beef, then this ends up dom­i­nat­ing veg*n out­reach in terms of re­duc­ing the num­ber of farmed an­i­mals.

  • As usual this means that if you think cows have net pos­i­tive lives and are sig­nifi­cantly sen­tient this may be net nega­tive. (As­sum­ing they are not re­placed by some­thing bet­ter!)

  • With de­fault val­ues this is also an ex­ceed­ingly effi­cient way of re­duc­ing CO2 emis­sions, at a cost of just 2 cents per tonne of CO2, which also makes it a great way of sav­ing QALYs, at $4 per QALY. If you are very scep­ti­cal about the value of x-risk re­search, this could also be the most cost-effec­tive way of re­duc­ing x-risk (via cli­mate change re­duc­tion).

  • Depends strongly on: the chance of cul­tured meat, chance of global catas­trophic and x-risk, an­i­mal qual­ity of life, im­por­tance of cli­mate change.


  • In­creased con­sump­tion slightly in­creases CO2, which then slightly re­duces pop­u­la­tion and in­creases x-risk.

  • Still ro­bustly pos­i­tive on short-term to­tal view due to in­creas­ing hap­piness, end­ing up with around $500 per QALY on de­fault val­ues.

  • To a small ex­tent in­creases the num­ber of farmed an­i­mals, which is bad if you think they have net-nega­tive lives, but this is swamped by the pos­i­tive effects on hu­mans un­der all weight­ings that value an­i­mals less than or equal to hu­mans (and no mat­ter how bad you as­sume an­i­mal lives are on our scale).

  • Seems ro­bustly good (ig­nor­ing far fu­ture effects) and not very sen­si­tive to any vari­ables.


  • Essen­tially just op­er­ates as a mul­ti­ple of GiveDirectly, de­pend­ing on the “effect of child­hood de­worm­ing on fu­ture in­come” vari­able, so the com­ments on GiveDirectly ap­ply here whole­sale.

  • For the de­fault val­ues it ends up be­ing around 30 times bet­ter than GiveDirectly, giv­ing $16 per hu­man QALY, due to the po­ten­tial of de­worm­ing mas­sively boost­ing fu­ture in­come.

  • Again this is en­tirely de­pen­dant on the “effect of de­worm­ing on life­time earn­ings” vari­able.

Against Malaria Foundation

  • Effec­tively sim­ply saves QALYs at AMF’s stan­dard rate, and very slightly in­creases pop­u­la­tion and wellbe­ing.

  • In­creases num­ber of an­i­mals be­ing farmed but by neg­ligible amounts.

  • QALY effects de­pend strongly on num­ber of QALYs given to pre­vent­ing the death of a child un­der 5, but oth­er­wise ro­bust.

EA Outreach

  • Not cur­rently well-im­ple­mented, ba­si­cally just acts as mul­ti­plier fund­ing other ar­eas.

  • Al­most all value with our de­fault set­tings comes from x-risk, as this gen­er­ally dom­i­nates.

  • Sen­si­tive to all vari­ables to differ­ent de­grees, with x-risk vari­ables hav­ing the biggest effect.

Cause prioritisation

  • Not cur­rently in­te­grated at all: ended up seem­ing like a sep­a­rate pro­ject to model well. While we came up with a model for cause pri­ori­ti­sa­tion, we needed bet­ter data on things like the un­der­ly­ing dis­tri­bu­tion of cause effec­tive­ness and re­searchers’ abil­ity to dis­cover cause effec­tive­ness.

  • This does how­ever seems po­ten­tially very im­por­tant, and we would like to see it stud­ied fur­ther.

Global catas­trophic and x-risk policy and strategy

  • Very strong x-risk re­duc­tion effects un­der de­fault val­ues (0.13% points re­duc­tion per mil­lion dol­lars) which leads to in­creased ex­pected pop­u­la­tion. This can be nega­tive in the to­tal view if you think most hu­mans have neu­tral or nega­tive lives.

  • Depends strongly on global catas­trophic and x-risk base chance, and how much gov­ern­ment policy can in­fluence x-risk.

  • Su­pe­ri­or­ity over other x-risk re­duc­tion paths is en­tirely de­pen­dant on very un­cer­tain (and quite ar­bi­trary) val­ues for how much policy effects x-risk etc.

Global catas­trophic and x-risk research

  • Medium x-risk re­duc­tion un­der de­fault val­ues (0.05% points re­duc­tion per mil­lion dol­lars), with same im­pli­ca­tions as above.

  • Depends strongly on base chance of global catas­trophic and x-risk, and how much re­search re­duces x-risk.

Far fu­ture (global catas­trophic and x-risk) outreach

  • Small x-risk re­duc­tion un­der de­fault val­ues (0.03% points re­duc­tion per mil­lion dol­lars), with same im­pli­ca­tions as above. Un­der de­fault as­sump­tions al­most all im­pact comes from the effects of gov­ern­ment policy, not re­search, but this is very un­cer­tain (as above).

  • Depends strongly on the base chance of global catas­trophic and x-risk, and how much how much gov­ern­ment policy can in­fluence x-risk.

Gen­eral an­i­mal out­reach /​ far fu­ture /​ global poverty

  • Th­ese nodes func­tion sim­ply by adding fund­ing to the spe­cific ar­eas in their cause, e.g. fund­ing Global Poverty just funds AMF, Givedi­rectly and De­worm­ing.

  • Their effects are just com­bi­na­tions of the effects of the nodes dis­cussed ear­lier, of­ten dom­i­nated by whichever area hap­pens to be largest. Dis­cus­sion of the im­pacts of fund­ing these ‘cause level’ nodes will be ex­cluded as we can look at the spe­cific ar­eas within causes for a more com­plete pic­ture.


Our next post will be a writeup of one par­tic­u­lar find­ing of the model, the cli­mate change catas­trophic risk con­nec­tion find­ings of the model which con­sti­tutes Part IV of the se­ries.

Feel free to ask ques­tions in the com­ment sec­tion, or email us ( or


[1]: Non-caged hens seem to emit about 15-25% more green­houses gases per egg (due to in­creased food, heat­ing and land re­quire­ments and a higher pro­por­tion of eggs be­ing lost), which re­sults in around 500,000 tonnes more CO2 emit­ted per year. Our very rough es­ti­mates lead to the con­clu­sion that ap­prox­i­mately each ad­di­tional 1,000 Tonnes of CO2e emit­ted will cause the loss of one QALY be­fore 2050.

[2]: De­fault as­sump­tions used in our model were 7% x-risk chance by 2050, and 10,000 re­searches work­ing for a decade could half x-risk to 3.5%.Hence 1 re­searcher year re­duces risk by 0.000035% (per­centage points) and say one re­searcher year costs $50,000, so 7*10^-10 per­centage points re­duc­tion in x-risk per $. If ex­tinc­tion costs 25*7 billion = 175 billion QALYS, mul­ti­ply­ing out gives just over 1 QALY saved per $.