Review of Climate Cost-Effectiveness Analyses

This post was prompted by the com­ments on my pro­posed up­dated 80K Hours Cli­mate Change Prob­lem Pro­file.

It’s im­por­tant to make it clear up front that the sur­pris­ing truth is that there is gen­uinely very lit­tle quan­ti­ta­tive re­search into the im­pacts of cli­mate change of 4C and above. The re­search which does ex­ist is nec­es­sar­ily limited in scope and makes a large num­ber of as­sump­tions—many of which will tend to un­der­value the over­all im­pact of cli­mate change.

In this post I ex­am­ine four pre­vi­ous at­tempts to ex­am­ine as­pects of the im­pact of cli­mate change and/​or the cost-effec­tive­ness of cli­mate change in­ter­ven­tions. Full de­tails of these analy­ses are in­cluded be­low, but their head­line figures are sum­marised here:

  1. 2016 GWWC es­ti­mate − 2.8C tem­per­a­ture in­crease by 2100 pro­duces mor­tal­ity es­ti­mates that [with pro­posed model fixes ap­plied by me] sug­gest Cool Earth can save a life for ~$6,000, com­pared to $3,461 to save a life with Against Malaria Foun­da­tion [with enor­mous un­cer­tainty about this es­ti­mate]

  2. 2018 Halstead Ex­tinc­tion Risk - <1% − 3.5% ex­tinc­tion risk (>10C of warm­ing)

  3. 2019 Bressler Mor­tal­ity Es­ti­mate − 4.1C tem­per­a­ture in­crease by 2100 re­sults in 76 mil­lion deaths [pro­vi­sional re­sults from an in-progress PhD]

  4. 2019 Hille­brandt Cost-Effec­tive­ness − 2.2C tem­per­a­ture in­crease by 2100 pro­duces a SCC that [with pro­posed model fixes ap­plied by me] sug­gests Cool Earth is 1.15x more effec­tive than global health in­ter­ven­tions [Range: 0.0000003x − 4,041x]

Based on cur­rently an­nounced na­tional com­mit­ments, green­house emis­sions are likely to lead to global tem­per­a­ture in­creases of 2.3ºC-3.7ºC by 2100 with a 25% chance of ex­ceed­ing 4°C based on cur­rent na­tional poli­cies. This sug­gests that (1) and (4) are un­der­valu­ing ac­tion on cli­mate change since they are based on much lower lev­els of pro­jected warm­ing. Fur­ther­more, (1) and (4) both have very large flaws in their method­ol­ogy which are likely to dra­mat­i­cally un­der-value cli­mate ac­tion—see be­low for full de­tails.

(3) pro­jects 76 mil­lion deaths over the pe­riod 2020-2100. This is of a similar mag­ni­tude to the to­tal deaths caused by the sec­ond world war (70-85 mil­lion peo­ple over 6 years). This is also of a similar mag­ni­tude to the largest famines seen in the 20th cen­tury (1-2M peo­ple/​year). Th­ese kinds of num­bers give an idea of the scale of im­pact which we can ex­pect if cli­mate change of 4C hap­pens.

(2) com­puted an ex­is­ten­tial risk of <1% − 3.5%. This risk is not ac­counted for in any of the ex­ist­ing cost-effec­tive­ness analy­ses which only fo­cus on the av­er­age case along with a high/​low es­ti­mate of im­pact.

One of the cen­tral ideas in effec­tive al­tru­ism is that some in­ter­ven­tions are or­ders of mag­ni­tude more effec­tive than oth­ers. There re­main huge un­cer­tain­ties and un­knowns which make any at­tempt to com­pute the cost effec­tive­ness of cli­mate change ex­tremely challeng­ing. How­ever, the es­ti­mates which have been com­pleted so far don’t make a com­pel­ling case that miti­gat­ing cli­mate change is ac­tu­ally or­der(s) of mag­ni­tude less effec­tive com­pared to global health in­ter­ven­tions, with many of the re­main­ing un­cer­tain­ties mak­ing it very plau­si­ble that cli­mate change in­ter­ven­tions are in­deed much more effec­tive.

More­over, this re­sult is reached when only con­sid­er­ing the im­pact of deaths at­tributed to cli­mate change. This seems like an enor­mously nar­row lens through which to con­sider a prob­lem which risks dis­plac­ing hun­dreds of mil­lions of peo­ple, threat­en­ing global food sys­tems, caus­ing mas­sive species ex­tinc­tion, and could trig­ger cli­mate tip­ping points that am­plify all of these pro­jected im­pacts. Given all of this, it seems ex­tremely likely that cli­mate change miti­ga­tion is ac­tu­ally at least an or­der of mag­ni­tude more cost-effec­tive than the best available global health in­ter­ven­tions.

1. Background

1.1. Discounting

Dis­count­ing of fu­ture val­ues is a com­mon prac­tice in eco­nomics which has a huge im­pact on the fore­cast im­pact of cli­mate change. Cli­mate change is already im­pact­ing the world to­day and, if emis­sions con­tinue, the im­pacts are ex­pected to con­tinue to get much worse. Many fore­casts only choose to con­sider im­pacts within the 21st cen­tury, and hence the worst of these im­pacts will be at the end of that pe­riod. Tak­ing a cou­ple of ex­em­plar years − 2050 and 2100 - these are roughly 30 and 80 years away. The im­pact of differ­ent lev­els of dis­count­ing is as fol­lows:

  • 1% → 30 years: 74%, 80 years → 45%

  • 2% → 30 years: 55%, 80 years → 20%

  • 3% → 30 years: 40%, 80 years → 9%

This means that if you choose to dis­count the fu­ture by 2%/​year, then you are choos­ing to value im­pacts in 2100 as only 20% as im­por­tant as if they were hap­pen­ing to­day. There­fore it’s im­por­tant to ask what level of dis­count­ing is be­ing ap­plied when you look at cli­mate im­pact fore­casts.

If you be­lieve that lives in the fu­ture are also valuable, per­haps even just as valuable as lives to­day, then you may choose a very low or even zero dis­count rate and this will have a very large im­pact on your re­sult­ing val­u­a­tion of cli­mate change im­pact.

1.2. Global Mortality

Some of the es­ti­mates be­low are ex­pressed in terms of num­ber of deaths per year. To put these num­bers in con­text, it’s use­ful to have a few points of com­par­i­son.

  • Globally, there are cur­rently ~60 mil­lion deaths/​year across all causes, in­clud­ing those re­lated to age re­lated deaths. This is fore­cast to grow to ~120 mil­lion deaths/​year by 2100 due to pop­u­la­tion growth and an ag­ing pop­u­la­tion [source]

  • In the 20th cen­tury, the largest famines kil­led 10-20M peo­ple/​decade, so 1-2M peo­ple/​year, all of which hap­pened when the world had fewer than 4 billion peo­ple [source]

  • Since the 1960s, wars have kil­led at most 300K peo­ple/​year [source]

  • World War II kil­led 70-85 mil­lion peo­ple over 6 years, which is 11.7-14.2 mil­lion peo­ple/​year, at a time when the world pop­u­la­tion was ~2.3 billion [source]

  • In 2017, 437K peo­ple died from Malaria [source]

1.3. IAM Val­idity Concerns

Two of the es­ti­mates be­low are based on In­te­grated Assess­ment Models (IAMs). Se­ri­ous con­cerns have been raised with the use of these mod­els.

“In a re­cent ar­ti­cle, I ar­gued that in­te­grated as­sess­ment mod­els (IAMs) “have cru­cial flaws that make them close to use­less as tools for policy anal­y­sis.” In fact, I would ar­gue that call­ing these mod­els “close to use­less” is gen­er­ous: IAM-based analy­ses of cli­mate policy cre­ate a per­cep­tion of knowl­edge and pre­ci­sion that is illu­sory, and can fool policy-mak­ers into think­ing that the fore­casts the mod­els gen­er­ate have some kind of sci­en­tific le­gi­t­i­macy. IAMs can be mis­lead­ing – and are in­ap­pro­pri­ate – as guides for policy, and yet they have been used by the gov­ern­ment to es­ti­mate the so­cial cost of car­bon (SCC) and eval­u­ate tax and abate­ment poli­cies. What are the cru­cial flaws that make IAMs so un­suit­able for policy anal­y­sis? They are dis­cussed in de­tail in Pindyck (2013b), but the most im­por­tant ones can be briefly sum­ma­rized as fol­lows:

1. Cer­tain in­puts – func­tional forms and pa­ram­e­ter val­ues – are ar­bi­trary, but have huge effects on the re­sults the mod­els pro­duce. An ex­am­ple is the dis­count rate. There is no con­sen­sus among economists as to the “cor­rect” dis­count rate, but differ­ent rates will yield wildly differ­ent es­ti­mates of the SCC and the op­ti­mal amount of abate­ment that any IAM gen­er­ates. For ex­am­ple, these differ­ences in in­puts largely ex­plain why the IAM based analy­ses of Nord­haus (2008) and Stern (2007) come to such strik­ingly differ­ent con­clu­sions re­gard­ing op­ti­mal abate­ment. Be­cause the mod­eler has so much free­dom in choos­ing func­tional forms, pa­ram­e­ter val­ues, and other in­puts, the model can be used to ob­tain al­most any re­sult one de­sires, and thereby le­gi­t­imize what is es­sen­tially a sub­jec­tive opinion about cli­mate policy.

2. We know very lit­tle about cli­mate sen­si­tivity, i.e., the tem­per­a­ture in­crease that would even­tu­ally re­sult from a dou­bling of the at­mo­spheric CO2 con­cen­tra­tion, but this is a key in­put to any IAM. The prob­lem is that the phys­i­cal mechanisms that de­ter­mine cli­mate sen­si­tivity in­volve cru­cial feed­back loops, and the pa­ram­e­ter val­ues that de­ter­mine the strength (and even the sign) of those feed­back loops are largely un­known, and are likely to re­main un­known for the fore­see­able fu­ture.

3. One of the most im­por­tant parts of an IAM is the dam­age func­tion, i.e., the re­la­tion­ship be­tween an in­crease in tem­per­a­ture and GDP (or the growth rate of GDP). When as­sess­ing cli­mate sen­si­tivity, we can at least draw on the un­der­ly­ing phys­i­cal sci­ence and ar­gue co­her­ently about the rele­vant prob­a­bil­ity dis­tri­bu­tions. But when it comes to the dam­age func­tion, we know vir­tu­ally noth­ing – there is no the­ory and no data that we can draw from.

4. IAMs can tell us noth­ing about the like­li­hood or pos­si­ble im­pact of a catas­trophic cli­mate out­come, e.g., a tem­per­a­ture in­crease above 5°C that has a very large im­pact on GDP. And yet it is the pos­si­bil­ity of a cli­mate catas­tro­phe that is (or should be) the main driv­ing force be­hind a stringent abate­ment policy.”

[Pindyck, 2017, The Use and Mi­suse of Models for Cli­mate Policy]

Fur­ther rele­vant crit­i­cism can be read in [Weitz­man, 2011, Fat-Tailed Uncer­tainty in the Eco­nomics of Catas­trophic Cli­mate Change].

2. Cli­mate Change Im­pact /​ Cost-Effec­tive­ness Estimates

2.1. 2016 Giv­ing What We Can Cost-Effectiveness

2.1.1. Approach

Giv­ing What We Can (GWWC) de­scribe their ap­proach and re­sults here. The ap­proach can be sum­marised as fol­lows.

  1. So­cial Cost of Car­bon (SCC) re­jected as an ap­pro­pri­ate mea­sure of im­pacts.

  2. WHO’s 2014 re­port “Quan­ti­ta­tive risk as­sess­ment of the effects of cli­mate change on se­lected causes of death, 2030s and 2050s” [source] se­lected as key source of mor­tal­ity es­ti­mates. This re­port es­ti­mates in­cre­men­tal cli­mate change re­lated mor­tal­ity in 2030 and 2050 for heat-re­lated mor­tal­ity; coastal flood mor­tal­ity; di­ar­rhoeal dis­ease; malaria; Dengue fever; and un­der­nu­tri­tion. The A1B emis­sions sce­nario is used which pre­dicts 2.8C tem­per­a­ture in­crease by 2100 [source].

  3. The es­ti­mates in 2030 and 2050 are as­sumed to define a lin­ear re­la­tion­ship be­tween year and num­ber of deaths. The cen­tral value is an in­crease of 201.2 ex­tra deaths/​year on top of a baseline of 241K/​year in­cre­men­tal deaths in 2030.

  4. The causes of deaths in the WHO re­port only ac­count for 5.1% of to­tal mor­tal­ity, so as a con­ser­va­tive es­ti­mate, all causes of death are as­sumed to scale by the same amount, so 201.2*(100/​5.1)=3931 ex­tra to­tal deaths/​year.

  5. Re­duc­ing emis­sions in a year de­lays some frac­tion of these ex­tra deaths/​year.

  6. Hence, the cost of an emis­sions re­duc­tion can be mul­ti­plied through to reach a cost per life saved. The cen­tral es­ti­mate is $97,300, and the most gen­er­ous es­ti­mate is $32,700.

  7. No dis­count­ing is ap­plied in the fi­nal re­ported figures, al­though the spread­sheet al­lows this to be added on at the end.

2.1.2. Comments

2.1.2.1. WHO Re­port Limitations

The pro­jec­tions in the WHO re­port [source] come with a large num­ber of limi­ta­tions and caveats. Many of these are de­scribed in the re­port it­self—the au­thors are clearly aware of the great difficulty in­volved in pro­duc­ing these kinds of es­ti­mates. How­ever, these limi­ta­tions are so se­vere that the re­sult­ing num­bers must be used with ex­treme cau­tion. Let’s con­sider a cou­ple of the sec­tions.

Malnutrition

Malnu­tri­tion mod­el­ing doesn’t ac­count for: in­creases in ex­treme weather events, sea-level rise, changes in wa­ter de­mand, in­creases in pests and dis­eases, loss of in­come from land which be­comes un­pro­duc­tive. [WHO 2014, p70]

“We be­lieve our es­ti­mates should be con­sid­ered very con­ser­va­tive for the fol­low­ing two rea­sons [...] our mod­el­ling does not in­clude the im­pact of shocks; it con­sid­ers stunt­ing due only to ex­pected av­er­age con­di­tions” [WHO 2014, p96]

Coastal Flooding

As­sump­tions:

  • There is no change in storm-surge fre­quency and in­ten­sity from baseline (but flood­wa­ters are deeper with sea-level rise).

  • Peo­ple flooded on av­er­age once a year au­tonomously leave the area and are not at risk of flood­ing and hence mortality

  • Sea level rise: Aver­age global warm­ing was 2.4°C by the 2050s and 3.8°C by the 2090s. This cor­re­sponds to global mean sea-level rises of 0.22 m by 2050 and 0.37 m by 2080.

This last as­sump­tion is clearly out of date with the IPCC fore­cast­ing 0.52m of sea level rise in a 1.5C of warm­ing world by 2100 - “Model-based pro­jec­tions of global mean sea level rise (rel­a­tive to 1986–2005) sug­gest an in­dica­tive range of 0.26 to 0.77 m by 2100 for 1.5°C of global warm­ing” [IPPC SR15 - Sum­mary for Poli­cy­mak­ers]

The GWWC es­ti­mate doesn’t use the coastal flood­ing mor­tal­ity es­ti­mates as the WHO re­port only fore­casts these within broad bands (e.g. 10K-30K) [WHO 2014, p35] and the es­ti­mates don’t turn out to change be­tween bands be­tween 2030 and 2080. Given more re­cent es­ti­mates of much greater sea level rise, this no longer seems plau­si­ble.

2.1.2.2. Lin­ear Assumption

The GWWC model re­lies on the as­sump­tion that the point es­ti­mates given for mor­tal­ity in 2030 and 2050 can be ex­trap­o­lated into a lin­ear re­la­tion­ship. This seems like a deeply flawed as­sump­tion which is con­trary to aca­demic work such as this 2015 na­ture pa­per—Global non-lin­ear effect of tem­per­a­ture on eco­nomic pro­duc­tion [slides]. It also fails some ba­sic san­ity tests as the pre­sented num­bers claim that cli­mate change is caus­ing ex­cess malaria and di­ar­rhoeal dis­ease deaths, but that as cli­mate change wors­ens, it causes fewer of these deaths.

2.1.2.3. Ex­pand­ing To All Causes Of Death

The GWWC model as­serts “we can quite roughly es­ti­mate that mor­tal­ity due to cli­mate change might grow pro­por­tion­ally to cur­rent lev­els of mor­tal­ity—that is, that these dis­eases which cur­rently make up 5.117% of global mor­tal­ity will make up 5.117% of ad­di­tional mor­tal­ity due to cli­mate change and, hence, that deaths due to cli­mate change are 19.54 times higher than es­ti­mated in the WHO’s as­sess­ment.”

This as­ser­tion is weak as the re­sult­ing es­ti­mate is de­pen­dent on the five es­ti­mates taken from the WHO re­port. The two largest terms are (1) “ex­ces­sive heat”, ris­ing at 2851 deaths/​year be­tween 2030 and 2050, from a baseline of 37K in 2030, (2) “malaria”, de­clin­ing at 1369 deaths/​year be­tween 2030 and 2050, from a baseline of 60K in 2030. If the GWWC es­ti­mate had not in­cluded Malaria (by choice, or if Malaria had not been in the 2014 WHO re­port), then the change in deaths/​year be­tween 2030 and 2050 would have risen from 201/​year to 1571/​year. There were 435K malaria deaths in 2017 [source], which is ~0.7% of global deaths. 1571*100/​(5.117-0.7) = 35.6K/​year, rather than the origi­nal 201/​year es­ti­mate. So the es­ti­mate of change in deaths/​year is very sen­si­tive to the choice of es­ti­mates to in­clude be­fore mul­ti­ply­ing out.

Fi­nally, it seems wrong to count pro­jected re­duc­tions in malaria deaths against cli­mate change ac­tion when the re­duc­tion in deaths is pre­sum­ably pri­mar­ily be­cause of di­rect ac­tion against malaria. If the cli­mate was not warm­ing, you would ex­pect malaria to be de­clin­ing more rapidly, but the GWWC model seems to im­ply the re­verse. In fact, there is a cam­paign to elimi­nate malaria by 2040 [source], that if suc­cess­ful, would fur­ther in­val­i­date the GWWC model which at­tributes malaria death re­duc­tions to cli­mate change un­til long af­ter this date.

2.1.2.4. Lives Are Saved Every Year

This ap­pears to be one of the biggest flaws with the GWWC es­ti­mate. The GWWC es­ti­mate works on the ba­sis that re­duc­ing emis­sions saves some frac­tion of the in­crease in deaths that would have hap­pened as a re­sult of those emis­sions. How­ever, this sav­ing ac­tu­ally ap­plies for ev­ery year af­ter the emis­sions were re­duced.

The world cur­rently emits 37Gt CO2/​year. Ig­nor­ing longer term CO2 ab­sorp­tion pro­cesses, as­sum­ing these emis­sions con­tinued at that rate in­definitely, if emis­sions are re­duced by 1Gt in one year, then at­mo­spheric CO2 con­cen­tra­tions will be lower ev­ery sub­se­quent year than they would have been oth­er­wise.

So the ques­tion is, how many years of saved lives should be in­cluded in the calcu­la­tion? In the­ory the cor­rect num­ber should be the time un­til a given emis­sion of CO2 has later been re­cap­tured and se­questered. We don’t ex­pect to be able to re­cap­ture most emit­ted CO2, so a very con­ser­va­tive value to use would be to at­tribute 50 years of in­creased deaths to each emis­sion. Hence, this in­creases the es­ti­mate of lives saved by a fac­tor of 50x. This also ig­nores any other im­pacts of a given CO2 emis­sion, some of which are ac­tu­ally or effec­tively ir­re­versible, such as trig­ger­ing cli­mate tip­ping points, species ex­tinc­tion, and sea level rise.

2.1.2.5. Use Of Cen­tral WHO Estimates

Cells C46 - E50 con­tain the es­ti­mates of lives saved for a given emis­sions re­duc­tion. Th­ese cells fol­low the same for­mat as the rest of the sheet, with a cen­tral, low, and high es­ti­mate. How­ever, these es­ti­mated are all based on the cen­tral WHO es­ti­mates. The only vari­a­tion comes from use of a (cen­tral, low, high) es­ti­mate for the cost per acre of land pro­tected by Cool Earth and the down­ward effect of adap­tion.

2.1.2.6. Han­dling Of Pro­jected Declines

In the ar­eas of Un­der­nu­tri­tion, Malaria, and Diar­rhoeal deaths, the WHO es­ti­mates showed de­clin­ing cli­mate change at­tributed mor­tal­ity be­tween 2030 and 2050. Cells C48-C50 re­verses the sign of these es­ti­mates, which means they add to the lives saved rather than sub­tract­ing from them. I can’t see any ra­tio­nale for this.

2.1.3. Up­dated Estimate

I have at­tempted to pro­duce an up­dated es­ti­mate with the fol­low­ing changes:

  • I have re­moved con­sid­er­a­tion of malaria deaths which may have been en­tirely elimi­nated by 2040 and have ad­justed the “Per­centage of to­tal deaths” figures down­ward by the ap­prox 0.7% of global deaths caused by malaria to­day.

  • Num­ber of lives saved are taken to be 50x the re­duc­tion in per-year in­cre­men­tal cli­mate at­tributed deaths.

  • I have up­dated the low/​high es­ti­mates to ac­tu­ally use the low/​high es­ti­mates of cli­mate mor­tal­ity.

  • Re­mov­ing sign re­ver­sal from Un­der­nu­tri­tion, and Diar­rhoeal deaths.

The re­sult­ing cen­tral es­ti­mate is $5,886 per life saved, which is the same or­der of mag­ni­tude as the $3,461 quoted to save a life by the Against Malaria Foun­da­tion.

The low and high es­ti­mates end up be­ing weird due to the method­ol­ogy used in the origi­nal es­ti­mate. For ex­am­ple, the low es­ti­mates for malnu­tri­tion are that in 2030 there are 119,807 fewer deaths, which drops to 29,203 fewer deaths by 2050. This pro­duces a “low” es­ti­mate that cli­mate change in­creases malnu­tri­tion mor­tal­ity by 4530 lives a year, com­pared to the me­dian es­ti­mate of cli­mate change re­duc­ing malnu­tri­tion mor­tal­ity by 524 lives a year. Th­ese kinds of num­bers lead to a re­vised range of cost per life saved of be­tween a low of $3,819/​life saved and a high of -$701/​life saved. This seems en­tirely non­sen­si­cal to me.

My up­dated model is available here.

2.2. 2018 Halstead Ex­tinc­tion Risk

2.2.1. Approach

Halstead posted to the EA fo­rums about his 2018 pa­per “Strato­spheric aerosol in­jec­tion re­search and ex­is­ten­tial risk”. This pa­per es­ti­mates the risk of hu­man ex­tinc­tion from cli­mate change by com­bin­ing the fol­low­ing es­ti­mates.

10C of warm­ing is cho­sen as the thresh­old above which cli­mate change will cause hu­man ex­tinc­tion.

Table 1 - At­mo­spheric CO2 Con­cen­tra­tion in 2100 → Probability

  • 400 − 1%

  • 500 − 5%

  • 600 − 20%

  • 700 − 30%

  • 800 − 20%

  • 900 − 15%

Note, the prob­a­bil­ities don’t sum to 100% - the 9% chance of >900 is ig­nored. The pa­per doesn’t ex­plain why.

Table 2 - Prob­a­bil­ity of warm­ing >10C, at each CO2 con­cen­tra­tion → Probability

  • 400 − 0.2%

  • 500 − 0.83%

  • 600 − 1.9%

  • 700 − 3.2%

  • 800 − 4.5%

  • 900 − 6.6%

De­duc­ing from the es­ti­mates in Tables 1 and 2, the un­con­di­tional prob­a­bil­ity of ex­is­ten­tial catas­tro­phe-level warm­ing is ∼3.5%. I use Weitz­man’s es­ti­mate of cli­mate sen­si­tivity be­cause it at­tempts to ac­count for cli­mate feed­backs which are im­por­tant from the point of view of ex­is­ten­tial risk re­duc­tion. How­ever, Weitz­man’s ECS es­ti­mate is highly con­tro­ver­sial, and there are a few rea­sons to think it may be too high. Nord­haus (2011a, 2011b) has crit­i­cised Weitz­man’s anal­y­sis of the sam­ple of IPCC model prob­a­bil­ity dis­tri­bu­tions across ECS. Weitz­man (2009a) has defended his ap­proach and noted that even if Nord­haus’ ap­proach is cor­rect, the prob­a­bil­ities in Table 2 would be re­duced by around 60%, which still sug­gests that the risk of ex­is­ten­tial catas­tro­phe is ∼1.5%.

[...]

Thus, the head­line es­ti­mate I have pro­duced in this sec­tion is highly con­tro­ver­sial and some lines of ar­gu­ment sug­gest that the ex­is­ten­tial risks of cli­mate change are (much) lower, plau­si­bly < 1%. This con­tro­versy should be borne in mind in what fol­lows.

[Halstead, 2018, p5]

So the range <1% − 3.5% is the ex­is­ten­tial risk pre­dicted by this pa­per.

2.2.2. Comments

The prob­a­bil­ities in the ta­bles above come from a 2011 Weitz­man pa­per “Fat-Tailed Uncer­tainty in the Eco­nomics of Catas­trophic Cli­mate Change”. This pa­per also in­cluded es­ti­mates of the prob­a­bil­ity of >5C of warm­ing.

Table 3 - Prob­a­bil­ity of warm­ing >5C, at each CO2 con­cen­tra­tion → Probability

  • 400 − 1.5%

  • 500 − 6.5%

  • 600 − 15%

  • 700 − 25%

  • 800 − 38%

  • 900 − 52%

Mul­ti­ply­ing this through in the same way as be­fore gives a 26.2% chance of greater than 5C of tem­per­a­ture in­crease. This is re­as­sur­ing as it is (very roughly) in line with the 25% chance of greater than 4C tem­per­a­ture in­crease pre­dicted here.

2.3. 2019 Bressler Mor­tal­ity Estimate

2.3.1. Approach

Bressler is a Sus­tain­able Devel­op­ment PhD can­di­date who is work­ing on ac­count­ing for cli­mate mor­tal­ity in the calcu­la­tion of the So­cial Cost of Car­bon (SCC). This work ex­tends William Nord­haus’ DICE In­te­grated Assess­ment Model (IAM). Bressler gave a pub­lic talk with some early re­sults from his work in July 2019 and posted it to the EA fo­rum.

This es­ti­mate is based on ex­am­in­ing how cli­mate change im­pacts global mor­tal­ity in a fu­ture with 4.1C of tem­per­a­ture in­crease by 2100. The model pre­dicts that over the next 80 years, 76 mil­lion cu­mu­la­tive ad­di­tional deaths are caused. Th­ese are deaths from health im­pacts, in­creased mur­der, and in­ter­group con­flict re­sponse. I have reached out to Bressler to find out more de­tails about what speci­fics are in­cluded/​ex­cluded from these es­ti­mates. Ac­count­ing for these deaths triples the SCC es­ti­mate. It should be noted that these are all pre­limi­nary num­bers.

The video shows a bar chart with the to­tal deaths in each 5 year pe­riod be­tween 2020 and 2100. The death rate is pro­jected to have reached 2.18 mil­lion deaths/​5 years by 2100.

2.4. 2019 Hille­brandt Cost-Effectiveness

2.4.1. Approach

Hille­brandt posted this es­ti­mate to the EA fo­rum in Oc­to­ber 2019. After post­ing, the es­ti­mate un­der­went a ma­jor up­date which changed the con­clu­sion. I will only be dis­cussing the up­dated ver­sion.

The model takes as an in­put an es­ti­mate of the SCC from a 2018 pa­per “Coun­try-level so­cial cost of car­bon” of US$417 per tonne of CO2 (66% CI: US$177–805). This is com­puted on the ba­sis of a 2% pure time prefer­ence dis­count­ing rate along with a 1.5% elas­tic­ity of marginal util­ity [See this for de­tails on growth-ad­justed dis­count­ing]. The pa­per uses RCP6.0 which is pro­jected to re­sult in 2.2C of warm­ing by 2100.

The SCC is then nor­mal­ised by the rel­a­tive util­ity of $1 in a poor coun­try ver­sus the US—us­ing a range of three val­ues (13,610x, 1,260x, 120x). The re­sult is mul­ti­plied fur­ther by a range of three val­ues for the rel­a­tive effec­tive­ness of the very best in­ter­ven­tions ver­sus di­rect cash trans­fers (17.5x, 7.95x, 0.83x). Fi­nally, the range of costs for offset­ting/​re­duc­ing emis­sions is taken to be ($232, $10, $0.02) based on a se­lec­tion of scal­able solu­tions.

The re­sult is that cli­mate change in­ter­ven­tions are pre­dicted to be X times as effec­tive than global de­vel­op­ment: (0.0000003x, 0.004x, 4,041x).

2.4.2. Comments

2.4.2.1. Use of IAM based SCC

As per sec­tion 1.3. there are se­ri­ous val­idity con­cerns with the IAM mod­els which un­derly es­ti­mates of the SCC. It’s un­clear to me whether these con­cerns ap­ply en­tirely to the 2018 pa­per un­der­ly­ing this anal­y­sis as it im­ple­ments its IAM differ­ently.

“we used coun­try-level cli­mate pro­jec­tions taken di­rectly from gridded en­sem­ble cli­mate model simu­la­tion data as well as coun­try-level eco­nomic dam­age rela-tion­ships taken di­rectly from em­piri­cal macroe­co­nomic analy­ses. As cli­mate and eco­nomic quan­tities are em­piri­cal in this anal­y­sis, these un­cer­tain­ties are prob­a­bil­is­tic in our out­put.”

The post on the EA fo­rum does note in Ap­pendix 3 that the val­idity of IAM mod­els are ques­tioned. How­ever, the me­dian SCC used from this pa­per of $477 is ac­tu­ally higher than other es­ti­mates which are of­ten in the $50-$200 range.

2.4.2.2. SCC Ex­cluded Costs

Ap­pendix 1 of the EA fo­rum post notes that the SCC used in the calcu­la­tion ex­cludes a num­ber of fac­tors which may turn out to be very im­por­tant, such as tip­ping points, ocean acid­ifi­ca­tion, sea level rise, and bio­di­ver­sity loss. This is used to jus­tify the use of a 10x higher SCC in the “pes­simistic” case.

2.4.2.3. Use of High Cost Per Tonne of CO2 Averted

The cost per tonne of CO2 averted is taken from a sam­ple of highly scal­able in­ter­ven­tions, with the low­est cost be­ing $0.02. The me­dian case is taken to be $10. How­ever, this seems like a sur­pris­ing choice given that an in­di­vi­d­ual choos­ing to donate their money to­wards cli­mate change to­day would definitely be able to find an in­ter­ven­tion which was cheaper than this. The 2016 GWWC es­ti­mate used a cost $0.38/​tonne for dona­tions to Cool Earth.

In the sec­tion about the cost of abate­ment, the fo­rum post quotes the v2.0 GHG abate­ment which was pub­lished in 2009 by McKinsey. The lat­est ver­sion is v2.1 from 2010, which is still very old at this point. A more re­cent pa­per from 2018, The Cost of Re­duc­ing Green­house Gas Emis­sions, com­putes an up­dated es­ti­mate. This pa­per says:

One sober­ing in­sight from the es­ti­mates in Table 2 is that many of the least-ex­pen­sive in­ter­ven­tions cover a small amount of CO2 re­duc­tions, whereas the scal­able tech­nolo­gies that are at the cen­ter of dis­cus­sions about a trans­for­ma­tion to a low-car­bon econ­omy—elec­tric ve­hi­cles, so­lar pho­to­voltaic pan­els, and offshore wind tur­bines—are among the most ex­pen­sive on the list.

How­ever, the pa­per goes on to ex­am­ine two case stud­ies of so­lar power and elec­tric cars and pro­poses that the ini­tially high costs come down dra­mat­i­cally with de­ploy­ment scale, and so us­ing to­day’s prices is mis­lead­ing.

Another dat­a­point to con­sider is the Draw­down Pro­ject, de­scribed on wikipedia

Pro­ject Draw­down is a cli­mate change miti­ga­tion pro­ject ini­ti­ated by Paul Hawken and cli­mate ac­tivist Amanda Joy Raven­hill. Cen­tral to the pro­ject is the com­pila­tion of a list of the “100 most sub­stan­tive solu­tions to global warm­ing.” The list, en­com­pass­ing only tech­nolog­i­cally vi­able, ex­ist­ing solu­tions, was com­piled by a team of over 200 schol­ars, sci­en­tists, poli­cy­mak­ers, busi­ness lead­ers and ac­tivists; The team mea­sured and mod­eled each solu­tion’s car­bon im­pact through the year 2050, its to­tal and net cost to so­ciety, and its to­tal life­time sav­ings.

The re­sults were pub­lished in a 2017 book and all the write­ups for the solu­tions are available on­line. The 80 solu­tions that it ex­am­ined that use es­tab­lished tech­nol­ogy, have an over­all cost/​tonne of $28.61. How­ever, the es­ti­mated sav­ings are $71.87/​tonne, for a net sav­ing/​tonne of $43.25. The sav­ings largely come from lower op­er­at­ing costs, so fi­nanc­ing will likely be re­quired to cover the ini­tial cap­i­tal costs of these solu­tions, which will in many cases pay for them­selves over time.

Fi­nally, a May 2019 EA fo­rum post pro­moted re­search by “Let’s Fund” which pro­moted fund­ing a think­tank to ad­vo­cate for in­creas­ing gov­ern­ment fund­ing for clean en­ergy R&D. The me­dian pro­jected fi­nan­cial re­turn calcu­lated by their fermi es­ti­mate was 28x.

2.4.3. Up­dated Estimate

I have pro­duced an up­dated es­ti­mate with the fol­low­ing as­sump­tions:

  • Me­dian SCC: $477 - no ad­just­ment for over/​un­der-estimation

  • In­come ad­just­ment: 120x—this is con­ser­va­tive about how much more valuable $1 is in a de­vel­op­ing country

  • Cost per tonne: $0.38 - this is taken from the 2016 GWWC estimate

  • GiveDirectly vs. global health in­ter­ven­tions: 7.95x—Me­dian Givewell char­ity effec­tive­ness vs. cash

On the ba­sis of these as­sump­tions, cli­mate change in­ter­ven­tion is 1.15x more effec­tive than global health in­ter­ven­tion.

There is clearly con­sid­er­able un­cer­tainty in this re­sult, given that the origi­nal es­ti­mate had a range of 0.0000003x − 4,041x, which is 10 or­ders of mag­ni­tude. How­ever, I claim that the ti­tle claim of the origi­nal EA fo­rum post, that “Global de­vel­op­ment in­ter­ven­tions are gen­er­ally more effec­tive than Cli­mate change in­ter­ven­tions” is far too strongly worded.

My up­dated model is available here.