How likely is a nuclear exchange between the US and Russia?


My pre­vi­ous posts ad­dress how bad a nu­clear war is likely to be, con­di­tional on there be­ing a nu­clear war (see this post on the deaths caused di­rectly by a US-Rus­sia nu­clear ex­change, and this post on the deaths caused by a nu­clear famine), but they don’t con­sider the like­li­hood that we ac­tu­ally see a US-Rus­sia nu­clear ex­change un­fold in the first place. In this post, I get a rough sense of how prob­a­ble a nu­clear war might be by look­ing at his­tor­i­cal ev­i­dence, the views of ex­perts, and pre­dic­tions made by fore­cast­ers. I find that, if we ag­gre­gate those per­spec­tives, there’s about a 1.1% chance of nu­clear war each year, and that the chances of a nu­clear war be­tween the US and Rus­sia, in par­tic­u­lar, are around 0.38% per year.

Pro­ject Overview

This is the fifth post in Re­think Pri­ori­ties’ se­ries on nu­clear risks. In the first post, I look into which plau­si­ble nu­clear ex­change sce­nar­ios should worry us most, rank­ing them based on their po­ten­tial to cause harm. In the sec­ond post, I ex­plore the make-up and sur­viv­abil­ity of the US and Rus­sian nu­clear ar­se­nals. In the third post, I es­ti­mate the num­ber of peo­ple that would die as a di­rect re­sult of a nu­clear ex­change be­tween NATO states and Rus­sia. In the fourth post, I es­ti­mate the sever­ity of the nu­clear famine we might ex­pect to re­sult from a NATO-Rus­sia nu­clear war. In this post, I get a rough sense of the prob­a­bil­ity of nu­clear war by look­ing at his­tor­i­cal ev­i­dence, the views of ex­perts, and pre­dic­tions made by fore­cast­ers. In the sixth and sev­enth posts, I es­ti­mate the di­rect and in­di­rect effects of nu­clear ex­changes be­tween (1) In­dia and Pak­istan and (2) China and its ad­ver­saries. Fu­ture work, to be pub­lished later in the sum­mer, will ex­plore the con­tra­dic­tory re­search around nu­clear win­ter, the im­pact of sev­eral nu­clear arms con­trol treaties, and the case for and against fund­ing par­tic­u­lar or­ga­ni­za­tions work­ing on re­duc­ing nu­clear risks.

Would the US and Rus­sia in­ten­tion­ally start a nu­clear war?

While a nu­clear war be­tween the US and Rus­sia by no means seems im­mi­nent, re­la­tions be­tween the two are very pre­car­i­ous, mainly as a re­sult of con­flict in Syria, the an­nex­a­tion of Crimea, and alle­ga­tions of Rus­sian in­terfer­ence in US poli­tics. Pres­i­dent Putin has made it clear that he is pes­simistic about the re­la­tion­ship be­tween the US and Rus­sia. In an in­ter­view ear­lier this week, he stated that “[re­la­tions with the US] are go­ing down­hill, they are get­ting worse and worse” — a sen­ti­ment that has no doubt been ex­ac­er­bated by US an­nounce­ment that it will de­ploy 1,000 US troops to Poland (i.e. right next to Rus­sia) (Reuters, 2019 June 13; see for ex­am­ple, BBC, 2019). Were any of these con­flicts to worsen sig­nifi­cantly, the ten­sions could es­ca­late to the point of war. Whether a war be­tween the US and Rus­sia would nec­es­sar­ily be­come nu­clear is un­clear. But it’s cer­tainly a risk.

On the other hand, nu­clear de­ter­rence is a com­pel­ling force. At least for now, both the US and Rus­sia main­tain nu­clear forces that are ‘sur­viv­able,’ mean­ing that both coun­tries have nu­clear weapons that are well-hid­den enough that the other couldn’t de­stroy them all in a first strike. This means that, if ei­ther coun­try de­cided to ini­ti­ate a first strike, it would al­most cer­tainly face nu­clear re­tal­i­a­tion. And, while both the US and Rus­sia seem more in­clined to tar­get each oth­ers’ nu­clear forces di­rectly rather than tar­get each other’s cities and in­dus­try, nei­ther coun­try could be sure that the other wouldn’t tar­get its cit­i­zens and econ­omy. Deter­rence the­o­rists con­clude that, as long as this is true, the two coun­tries will re­main in a Nash equil­ibrium — nei­ther coun­try be­ing will­ing to risk the worst con­se­quences of a nu­clear war.[1] The Brook­ings In­sti­tu­tion puts it this way: “no sane ad­ver­sary would be­lieve that any poli­ti­cal or mil­i­tary ad­van­tage would be worth a sig­nifi­cant risk of the de­struc­tion of his own so­ciety” (Wirtz, 2000).

Un­for­tu­nately, it’s un­clear how sane — or rather, ra­tio­nal — the world lead­ers with the power to launch nu­clear weapons are. It’s easy to imag­ine that a world leader with few checks on their power would launch a nu­clear at­tack with­out suffi­ciently weigh­ing the grav­ity of the risks. And some lead­ers have more lee­way to launch a nu­clear first strike against an ad­ver­sary than oth­ers. For ex­am­ple, nei­ther Pres­i­dent Trump nor Kim Jong Un would have to get ap­proval from any­one else to launch a nu­clear weapon. Ac­cord­ing to the pres­i­dent of the Ploughshares Fund, “[t]he pres­i­dent can or­der a nu­clear strike in about the time it takes to write a tweet.”[2]

But it’s re­ally hard to know whether seem­ingly trig­ger-happy lead­ers would go through with launch­ing a first strike. Don­ald Trump has a rep­u­ta­tion for ex­ag­ger­a­tion and in­timi­da­tion (for ex­am­ple, see tweet be­low), and mil­i­tary ad­vi­sors around him have stated pub­li­cly that they would strongly ad­vise him not to launch a nu­clear at­tack if his de­ci­sion seemed ir­ra­tional (for ex­am­ple, see Diaz 2017). There’s also a strate­gic ad­van­tage gained by over-stat­ing the ex­tent to which one would se­ri­ously con­sider a nu­clear first strike. So we can’t ex­actly take world lead­ers at their word.

Over­all, my in­tu­ition is that de­ter­rence is effec­tive — that the promise of mu­tu­ally as­sured de­struc­tion is a strong enough dis­in­cen­tive that the US and Rus­sia would not risk their econ­omy and tens of mil­lions of cit­i­zens’ lives. But I could very well be un­der­es­ti­mat­ing the po­ten­tial for lead­ers to mis­judge the risks to their so­cieties and them­selves — or their po­ten­tial to not con­sider the risks at all.

But let’s say world lead­ers are suffi­ciently de­terred by the prospect of mu­tu­ally as­sured de­struc­tion, and wouldn’t be will­ing un­der al­most any cir­cum­stances to start a nu­clear ex­change. Might they start one by ac­ci­dent?

Ac­ci­den­tal nu­clear war

Since nu­clear weapons were de­vel­oped, there have been a num­ber of ac­ci­dents, equip­ment malfunc­tions, and hu­man mis­takes that brought the world close to nu­clear war. Some of these close calls were closer than oth­ers.

For ex­am­ple, in 1961, a B-52 bomber car­ry­ing two nu­clear war­heads broke apart in mid-air over Golds­boro, North Carolina (BBC, 2013). The breakup of the plane trig­gered the mechanism (a sim­ple pul­ley sys­tem) meant to be used by the pi­lot to drop the bomb over a tar­get, which started the arm­ing pro­cess for both bombs. One of the bombs landed safely on the ground, fail-safe in­tact, af­ter the suc­cess­ful de­ploy­ment of its parachute.

The other plum­meted to the ground af­ter its parachute failed and com­pleted three of its four arm­ing steps. Had it deto­nated, the bomb — which had over 250 times the ex­plo­sive yield of the nu­clear bomb dropped on Hiroshima — could have kil­led mil­lions of peo­ple across the East Coast (Mosher, 2017; Pao­letti, 2017).[3]

“The simu­lated blast ra­dius (small cir­cle) and fal­lout zone (wider bands) of a 3.8-mega­ton deto­na­tion in Faro, North Carolina.” (Pao­letti, 2017)

Ac­cord­ing to a now-de­clas­sified re­port on the event, now known as the Golds­boro in­ci­dent, “one sim­ple, dy­namo-tech­nol­ogy, low voltage switch stood be­tween the United States and a ma­jor catas­tro­phe” (The Guardian, 2013, p.1). The re­port later says that the “bomb did not pos­sess ad­e­quate safety for the air­borne alert role in the B-52” (The Guardian, 2013, p. 2). In an­other re­cently de­clas­sified re­port, Sec­re­tary of Defense McNa­mara is quoted as say­ing that only “by the slight­est mar­gin of chance, liter­ally the failure of two wires to cross, a nu­clear ex­plo­sion was averted” (McNa­mara (n. d.), p.1).

Other in­ci­dents like this are similarly alarm­ing. In 1960, early warn­ing sys­tems re­ported that long-range Rus­sian mis­siles had been launched with “99.9% cer­tainty” (Baum, de Neufville, & Bar­rett, 2018, 30). It turned out it was just the moon­rise, which officers even­tu­ally figured out af­ter no other warn­ing sys­tems were sound­ing alarms.

Two decades later, a tech­nolog­i­cal glitch caused early warn­ing sys­tems to go off at the US Strate­gic Air Com­mand, alert­ing offi­cials of an in­com­ing at­tack — mis­tak­enly (Baum, de Neufville, & Bar­rett, 2018). Strate­gic bomber pi­lots pre­pared to take off from their air bases, but the alert was found to be a false alarm when no other warn­ing sys­tems de­tected in­com­ing mis­siles.

A list of eigh­teen close calls caused by ac­ci­dents like this is available in Baum, de Neufville, and Bar­rett, (2018).

Given this his­tory of near-miss events, it’s not hard to imag­ine that hu­man or tech­nolog­i­cal er­rors could lead us to nu­clear war.

It’s tempt­ing to imag­ine that the risk of ac­ci­den­tal nu­clear war has de­creased as the tech­nolo­gies that have failed in the past, like early warn­ing radar and fail-safes have likely im­proved. I found some ev­i­dence of this, though not as much as I ex­pected. For ex­am­ple, the US’s early warn­ing sys­tems that de­tect in­com­ing mis­siles (which have been par­tially re­spon­si­ble for near-miss events in the past) have been up­graded with tech­nol­ogy that is bet­ter at clas­sify­ing and track­ing pro­jec­tiles (Owens, 2017). This likely re­duces the risk that the US or Rus­sia would mis-iden­tify me­te­o­rolog­i­cal events or non-nu­clear pro­jec­tiles as an im­pend­ing nu­clear at­tack.

But the ex­panded use of tech­nol­ogy in nu­clear weapons sys­tems may in­tro­duce_ new_ risks. Unal and Lewis warn that, be­cause nu­clear weapons now rely more heav­ily on digi­tal tech­nol­ogy, they’ve be­come more vuln­er­a­ble to cy­ber at­tack (Unal & Lewis, 2018).

It’s hard to guess the net effect of these de­vel­op­ments, in part be­cause we can’t be con­fi­dent that we know about all of the close calls that have hap­pened — es­pe­cially in the re­cent past (re­cent enough that the de­tails haven’t been de-clas­sified) (Lewis et al., 2014).[4]

Put­ting a num­ber on the prob­a­bil­ity of a US-Rus­sia nu­clear war

To get a rough sense of how prob­a­ble we should find an in­ten­tional or ac­ci­den­tal nu­clear war, I looked at his­tor­i­cal ev­i­dence, the views of ex­perts, and pre­dic­tions made by fore­cast­ers.

His­tor­i­cal evidence

We can es­tab­lish a base rate for the prob­a­bil­ity of nu­clear war by look­ing at the num­ber of times nu­clear weapons have been used dur­ing a war: one time since they were de­vel­oped 74 years ago. This could be in­ter­preted to mean that the like­li­hood of nu­clear war is around 1.4% per year.

But there are sev­eral rea­sons not to put much stock in this prob­a­bil­ity (Baum, de Neufville, & Bar­rett, 2018). The most im­por­tant rea­son, in my view, is the fact that the one time that nu­clear weapons were used dur­ing a war — by the US, dur­ing World War II — the US was the only coun­try who had nu­clear weapons. Be­cause of this, they weren’t de­terred by the threat of nu­clear re­tal­i­a­tion. To­day, be­cause coun­tries con­sid­er­ing us­ing nu­clear weapons risk a nu­clear sec­ond strike, the thresh­old for us­ing nu­clear weapons is likely higher than it was when the US used nu­clear bombs dur­ing WWII, thereby de­creas­ing the prob­a­bil­ity of nu­clear war some­what. \

Baum et al. (2018) also point out that a num­ber of other his­tor­i­cal cir­cum­stances have changed: the poli­ti­cal lead­ers, the num­ber of coun­tries with nu­clear weapons, and the re­la­tion­ships be­tween coun­tries with nu­clear weapons, among other things. And again, they note that the prob­a­bil­ity might be some­what higher than the base rate would sug­gest, given that there have been a num­ber of geopoli­ti­cal crises that al­most es­ca­lated to the point of nu­clear war but didn’t (near misses in grey box in the figure be­low):

Source: Adapted from Baum, de Neufville, and Bar­rett (2018)

To take these close calls into ac­count, re­searchers Bar­rett, Baum and Hostetler (2013) took an al­ter­na­tive ap­proach to es­ti­mat­ing the baseline prob­a­bil­ity of nu­clear war us­ing his­tor­i­cal fre­quen­cies (though they fo­cus only on the risk of ac­ci­den­tal nu­clear war be­tween the US and Rus­sia). Rather than look­ing only at the in­stances where nu­clear war ac­tu­ally hap­pened, they looked back at the fre­quency of ac­ci­dents that nearly led to nu­clear war. They ar­gued that, if one as­sumes that those near misses could have also ended in dis­aster, one can make in­fer­ences about the prob­a­bil­ity of nu­clear war that are more in­for­ma­tive than just look­ing at the base rate of nu­clear war.

Based on this rea­son­ing, Bar­rett et al. (2013) con­cluded that the me­dian an­nual prob­a­bil­ity of in­ad­ver­tent nu­clear war be­tween the US and Rus­sia is about 0.9% (90% CI: 0.02% — 7%).

But there’s con­tro­versy over how to in­ter­pret those close calls. Should we con­sider them ev­i­dence that a nu­clear war could eas­ily have hap­pened mul­ti­ple times since WWII — that we’re just lucky they didn’t? Or should we think of them as ev­i­dence that, while near-misses are rel­a­tively com­mon, it’s ac­tu­ally re­ally hard for a close call to es­ca­late to the point of a nu­clear ex­change?

It’s difficult to know for sure, but I’m struck by the fact that a num­ber of the close calls caused by ac­ci­dents re­ported by Baum et al. (2018) have similar-sound­ing end­ings — usu­ally some­thing like “be­cause there was no other ev­i­dence of an at­tack…” X agency “de­ter­mined it was a false alarm caused by” X malfunc­tion (Baum et al., 2018, p. 30). This could be in­ter­preted to mean that, even though hu­man and tech­nolog­i­cal er­ror may lead to more close calls that we’d hope, the sys­tems in place to iden­tify mis­takes be­fore they es­ca­late might just work well enough to keep nu­clear war from hap­pen­ing by ac­ci­dent. But this is only very weak ev­i­dence, and there’s a lot of rea­son to be un­cer­tain.

Given this un­cer­tainty, it’s use­ful to look to other forms of ev­i­dence for how likely a US-Rus­sia nu­clear ex­change — in­ten­tional or oth­er­wise — might be.

Ev­i­dence from ex­pert surveys

The 2008 Global Catas­trophic Risk (GCR) sur­vey asked ex­perts to make pre­dic­tions about nu­clear war sce­nar­ios through to 2100. Speci­fi­cally, the ex­perts were asked to es­ti­mate the like­li­hood that nu­clear wars kill a) at least one mil­lion peo­ple, b) at least one billion peo­ple, and c) enough peo­ple that hu­mans be­come ex­tinct (Sand­berg & Bostrom, 2008).

From this we can glean that ex­perts see the prob­a­bil­ity of nu­clear war kil­ling at least 1 mil­lion or 1 billion peo­ple by 2100 as rea­son­ably small, but not in­signifi­cant — about 0.39% and 0.12% per year, re­spec­tively.[5] They see the risk of ex­tinc­tion caused by nu­clear war as much smaller, but again, it’s not in­signifi­cant at about 0.011% per year.

Re­spon­dents weren’t asked to con­sider the prob­a­bil­ity of spe­cific nu­clear ex­change sce­nar­ios, so these figures can in­form our un­der­stand­ing of the like­li­hood of a nu­clear ex­change be­tween the US and Rus­sia.

Another ex­pert sur­vey, the Lu­gar Sur­vey On Pro­lifer­a­tion Threats and Re­sponses, asked ex­perts all over the world to es­ti­mate the prob­a­bil­ity of nu­clear at­tack, but over a shorter time span (Lu­gar, 2005).

The me­dian view of ex­perts es­ti­mat­ing the prob­a­bil­ity of a nu­clear at­tack within 5 years (from 2004-2009) was 10%, or 2.09% per year, and 20% over 10 years (from 2004-2014), or 2.21% per year.

Like the GCR sur­vey, the Lu­gar Sur­vey didn’t ask ex­perts to con­sider spe­cific geopoli­ti­cal sce­nar­ios, so again, we can only learn a limited amount about a US-Rus­sia ex­change. Ad­di­tion­ally, it’s worth not­ing that the five and ten-year time hori­zons re­flected in these pre­dic­tions have already passed and re­solved in the nega­tive.

I did iden­tify one sur­vey that asks ex­plic­itly about the prob­a­bil­ity of a nu­clear ex­change be­tween the NATO states and Rus­sia. The Pro­ject for the Study of the 21st Cen­tury (PS21) Great Power Con­flict Re­port, re­leased for pub­li­ca­tion in 2015, asked 50 na­tional se­cu­rity ex­perts from all over the world to es­ti­mate the prob­a­bil­ity of a va­ri­ety of con­flict sce­nar­ios that could plau­si­bly oc­cur in the next 20 years (Apps, 2015).[6][7] The me­dian view of the ex­perts was that the prob­a­bil­ity of a nu­clear ex­change be­tween the US and Rus­sia in the next 20 years is around 4.72%, or 0.24% per year.

Ev­i­dence from Good Judg­ment Inc. Superforecasts

Fi­nally, I con­sider ev­i­dence from Good Judg­ment Inc. (GJI). Good Judg­ment Inc. uses in­sights learned from the Good Judg­ment Pro­ject, a re­search ini­ti­a­tive that worked to un­der­stand how to make ac­cu­rate pre­dic­tions about the fu­ture by hold­ing fore­cast­ing tour­na­ments. Dur­ing these tour­na­ments, fore­cast­ers were asked to make pre­dic­tions about the fu­ture, and GJP re­searchers worked to un­der­stand what made some peo­ple perform bet­ter than oth­ers.

In 2018, GJI had its su­perfore­cast­ers make a set of pre­dic­tions about the prob­a­bil­ity of a nu­clear ex­change by the year 2021 (un­pub­lished GJI data from Open Philan­thropy Pro­ject). The GJI fore­casts are likely to be less sus­cep­ti­ble to the bi­ases that lead ex­perts to make some­what worse pre­dic­tions (AI Im­pacts, 2019).

The su­perfore­cast­ers made two rele­vant pre­dic­tions:

  1. The fore­cast­ers pre­dicted that there is a 1% chance of a nu­clear at­tack by a state ac­tor caus­ing at least one fatal­ity be­fore 1 Jan­uary 2021 — equiv­a­lent to about 0.40% per year.

  2. They then pre­dicted that, as­sum­ing (1) proves true, there is a 3% chance that a nu­clear at­tack by a state ac­tor causes at least one fatal­ity in Rus­sia dur­ing the same time span.

If most of the prob­a­bil­ity of a nu­clear at­tack on Rus­sia comes from a nu­clear ex­change be­tween the US and Rus­sia, then we can ap­prox­i­mate the prob­a­bil­ity of a US-Rus­sia nu­clear ex­change by calcu­lat­ing the prob­a­bil­ity that there is 1) a nu­clear at­tack by a state ac­tor and 2) a nu­clear at­tack by a state ac­tor in Rus­sia, which is 0.03%, or 0.01% per year (un­pub­lished GJI data from Open Philan­thropy Pro­ject; Apps, 2015).

It should be noted that the GJI pre­dic­tions were made with a very short time span and a spe­cific set of poli­ti­cal cir­cum­stances in mind; they can only re­ally tell us about the prob­a­bil­ity of a US-Rus­sia nu­clear ex­change be­fore 2021.

So, how likely is a nu­clear war be­tween the US and Rus­sia?

If we ag­gre­gate his­tor­i­cal ev­i­dence, the views of ex­perts and pre­dic­tions made by fore­cast­ers, we can start to get a rough pic­ture of how prob­a­ble a nu­clear war might be.[8] We shouldn’t put too much weight on these es­ti­mates, as each of the data points feed­ing into those es­ti­mates come with se­ri­ous limi­ta­tions. But based on the ev­i­dence pre­sented above, we might think that there’s about a 1.1% chance of nu­clear war each year and that the chances of a US-Rus­sia nu­clear war may be in the bal­l­park of 0.38% per year (but note that these figures are de­cep­tively pre­cise).

(See notes on sources[9][10][11][12] and use of es­ti­mates[13][14][15].)


June 12, 2019 — I in­cluded an in­cor­rect table when I meant to pre­sent a table sum­ma­riz­ing the re­sults of the Lu­gar Sur­vey On Pro­lifer­a­tion Threats and Re­sponses. I’ve re­placed that table with the cor­rect one.

June 12, 2019 — I in­cor­rectly at­tributed fore­casts to Good Judg­ment Pro­ject su­perfore­cast­ers when they were ac­tu­ally made by Good Judg­ment Inc.

June 12, 2019 — I origi­nally re­ported that the av­er­age Good Judg­ment Inc. su­perfore­caster pre­dic­tion in an­swer to the ques­tion “What is the prob­a­bil­ity that a nu­clear at­tack by a state ac­tor causes at least one fatal­ity be­fore 1 Jan­uary 2021” was 2%. When I re-calcu­lated this figure in­de­pen­dently us­ing dis-ag­gre­gated data pro­vided by Open Philan­thropy Pro­ject, I found that the av­er­age GJI pre­dic­tion as 1%. Ac­count­ing for this new es­ti­mate causes the av­er­age an­nu­al­ized prob­a­bil­ity of nu­clear war to fall from 1.17% to 1.1%, and the an­nu­al­ized prob­a­bil­ity of a US-Rus­sia nu­clear war to fall from 0.39% to 0.38%.


This es­say is a pro­ject of Re­think Pri­ori­ties. It was writ­ten by Luisa Ro­driguez with con­tri­bu­tions from Ida Sprengers. Thanks to Peter Hur­ford, Marinella Capriati, and Mar­cus A. Davis who pro­vided valuable com­ments. Thanks also to Matt Gentzel and Seth Baum for pro­vid­ing guidance and feed­back on the larger pro­ject, and to Carl Schul­man for sup­ply­ing some of the data re­ported in this post.

If you like our work, please con­sider sub­scribing to our newslet­ter. You can see all our work to date here.


AI Im­pacts. (2019, Fe­bru­ary 15). Ev­i­dence on good fore­cast­ing prac­tices from the Good Judg­ment Pro­ject: An ac­com­pa­ny­ing blog post. Retrieved from https://​​aiim­​​ev­i­dence-on-good-fore­cast­ing-prac­tices-from-the-good-judg­ment-pro­ject-an-ac­com­pa­ny­ing-blog-post/​​

Apps, P. (2015). PS21 Great Power Con­flict Re­port (Rep.). Pro­ject for the Study of the 21st Cen­tury. Retrieved from https://​​​​doc­u­ment/​​289407938/​​PS21-Great-Power-Con­flict-Report

Bar­rett, A. M., Baum, S. D., & Hostetler, K. (2013). An­a­lyz­ing and re­duc­ing the risks of in­ad­ver­tent nu­clear war be­tween the United States and Rus­sia. Science and Global Se­cu­rity. https://​​​​10.1080/​​08929882.2013.798984

Baum, S., de Neufville, R., & Bar­rett, A. (2018). A model for the prob­a­bil­ity of nu­clear war. So­cial Science Re­search Net­work (SSRN). https://​​​​10.2139/​​ssrn.3137081

Diaz, D. (2017, Novem­ber 20). Top gen­eral says he’d push back against ‘ille­gal’ nu­clear strike or­der. CNN. Retrieved from https://​​​​2017/​​11/​​18/​​poli­tics/​​air-force-gen­eral-john-hyten-nu­clear-strike-don­ald-trump/​​in­dex.html

80,000 Hours. (2018, De­cem­ber 27). We could feed all 8 billion peo­ple through a nu­clear win­ter. Dr David Denken­berger is work­ing to make it prac­ti­cal [Au­dio blog in­ter­view]. Retrieved from https://​​​​pod­cast/​​epi­sodes/​​david-denken­berger-al­lfed-and-feed­ing-ev­ery­one-no-mat­ter-what/​​

Golds­boro re­vis­ited: ac­count of hy­dro­gen bomb near-dis­aster over North Carolina – de­clas­sified doc­u­ment. (2013). The Guardian. Retrieved from https://​​www.the­​​world/​​in­ter­ac­tive/​​2013/​​sep/​​20/​​golds­boro-re­vis­ited-de­clas­sified-document

[Good Judg­ment Inc Su­perfore­casts]. (n.d.). Un­pub­lished raw data from Open Philan­thropy Pro­ject.

Lewis, P., Willi­ams, H., Pelop­i­das, B., Agh­lani, S. (2014, April). Too close for com­fort: Cases of near nu­clear use and op­tions for policy. Chatham House. Retrieved from https://​​​​sites/​​de­fault/​​files/​​field/​​field_doc­u­ment/​​20140428TooClose­forCom­fortNu­cle­arUseLewisWilli­am­sPelop­i­dasAgh­lani.pdf

Lu­gar, R. G. (2005). The Lu­gar Sur­vey on pro­lifer­a­tion threats and re­sponses. _United States Se­na­tor for In­di­ana Chair­man, Se­nate For­eign Re­la­tions Com­mit­tee. _Retrieved from http://​​­cle­​​menu/​​key-is­sues/​​nu­clear-weapons/​​is­sues/​​pro­lifer­a­tion/​​fuel-cy­cle/​​sen­ate-dot-gov_NPSur­vey.pdf

McNa­mara: Un­ti­tled Note. (n. d.). Retrieved from https://​​www.doc­u­ment­​​doc­u­ments/​​799655-mc­na­mara.html#doc­u­ment/​​p2/​​a122657

Mosher, D. (2017, May 7). A ther­monu­clear bomb slammed into a North Carolina farm in 1961 — and part of it is still miss­ing. Busi­ness­in­sider. Retrieved from https://​​www.busi­ness­in­​​nu­clear-bomb-ac­ci­dent-golds­boro-nc-swamp-2017-5

Os­born, A. & Tsvetkova, M. (2019, June 12). Putin says U.S.-Rus­sia re­la­tions are get­ting ‘worse and worse’. Reuters. Retrieved from https://​​​​ar­ti­cle/​​us-usa-rus­sia-putin/​​putin-says-rus­sia-u-s-re­la­tions-are-get­ting-worse-and-worse-mir-tv-idUSKCN1TE0L7

Owens, K. (2017, April 24). The Air Force up­grades its mis­sile early warn­ing sys­tems. Retrieved from https://​​defens­esys­​​ar­ti­cles/​​2017/​​04/​​24/​​ear­ly­warn­ingradar.aspx

Rus­sia marks five years since an­nex­a­tion of Ukraine’s Crimea, (2019, March 18). Al­jazeera. Retrieved from https://​​­​​news/​​2019/​​03/​​rus­sia-marks-years-an­nex­a­tion-ukraine-crimea-190318072208651.html

Sand­berg, A., & Bostrom, N. (2008). Global Catas­trophic Risks Sur­vey. Fu­ture of Hu­man­ity In­sti­tute, Oxford Univer­sity. Retrieved from http://​​­​​docs/​​2008-1.pdf

Syria’s war: Who is fight­ing and why (2017, April 7). Vox. Retrieved from https://​​​​watch?v=JFpanWNgfQY

Trump: US to send 1,000 troops to Poland in new deal. (2019, June 12). BBC. Retrieved from https://​​​​news/​​world-us-canada-48585045

Unal, B. & Lewis, P. (2018). Cy­ber­se­cu­rity of nu­clear weapons sys­tems: Threats, vuln­er­a­bil­ities and con­se­quences. Chatham House. Retrieved from https://​​​​sites/​​de­fault/​​files/​​pub­li­ca­tions/​​re­search/​​2018-01-11-cy­ber­se­cu­rity-nu­clear-weapons-unal-lewis-fi­nal.pdf

US plane in 1961 ‘nu­clear bomb near-miss’. (2013, Sept 21). BBC. Retrieved from https://​​​​news/​​world-us-canada-24183879

Van­den Brook, T. and Col­lins, M. (2019, April 19). Muel­ler re­port: 5 things to know about Rus­sian in­terfer­ence in U.S. elec­tions. USA To­day. Retrieved from https://​​­ato­​​story/​​news/​​poli­tics/​​2019/​​04/​​22/​​muel­ler-re­port-what-know-rus­sian-elec­tion-in­terfer­ence/​​3538877002/​​

Ward, A. (2018, De­cem­ber 26). This is ex­actly how a nu­clear war would kill you. Vox. Retrieved from https://​​​​fu­ture-perfect/​​2018/​​10/​​19/​​17873822/​​nu­clear-war-weapons-bombs-how-kill

What is the Nash equil­ibrium and why does it mat­ter? (2016, Septem­ber 17). The Economist. Retrieved from https://​​​​the-economist-ex­plains/​​2016/​​09/​​06/​​what-is-the-nash-equil­ibrium-and-why-does-it-matter

Wirtz, J. J. (2000). The Nu­clear Turn­ing Point: A Blueprint for Deep Cuts and De-Alert­ing of Nu­clear Weapons ed­ited by Harold A. Feive­son . Poli­ti­cal Science Quar­terly. https://​​​​10.2307/​​2658039

  1. See for ex­am­ple The Economist, 2016 ↩︎

  2. (Ward, 2018) ↩︎

  3. “If the bombs had deto­nated, lethal fal­lout could have been de­posited over Wash­ing­ton, Bal­ti­more, Philadelphia, and even as far north as New York City, putting mil­lions of lives at risk” (Pao­letti, 2017). ↩︎

  4. “It is im­pos­si­ble to say whether the risk of near misses has in­creased over time. This is pri­mar­ily be­cause it is not pos­si­ble to have a com­plete sense of the num­ber of near misses and there­fore to de­ter­mine if and when there was a greater con­cen­tra­tion of them. While the con­se­quences of a nu­clear deto­na­tion have re­mained rel­a­tively con­sis­tent, lev­els of risk prob­a­bil­ity are difficult to es­ti­mate. It took decades to learn about the role of mis­per­cep­tions in the Cuban mis­sile crisis, and there is no rea­son to as­sume that the full pic­ture has been drawn” (Lewis et al., 2014, p. 30) ↩︎

  5. Note, to es­ti­mate this an­nu­al­ized prob­a­bil­ity (and all other an­nu­al­ized prob­a­bil­ities in this post), I as­sume that each the prob­a­bil­ity of nu­clear war hap­pen­ing in a given year is in­de­pen­dent of the prob­a­bil­ity of nu­clear hap­pen­ing in any other year, which isn’t ac­tu­ally the case. ↩︎

  6. I be­lieve the Pro­ject for the Study of the 21st Cen­tury (PS21) Great Power Con­flict Re­port has sev­eral ty­pos (Apps, 2015). I pre­sent what I be­lieve to be the cor­rect val­ues (and the val­ues I use in my anal­y­sis) here. ↩︎

  7. Another typo in the PS21 Great Power Con­flict Re­port makes it un­clear whether the prob­a­bil­ities are over a 20-year time span or a 25-year time span. I as­sume the sur­vey asked about the time-span noted in a table, not in the dis­cus­sion of the re­sults. ↩︎

  8. A bet­ter way to ag­gre­gate these figures would be to use for­mal Bayesian up­dat­ing. Be­cause of time con­straints, I just take the sim­ple av­er­age in­stead, which can still be some­what in­for­ma­tive. ↩︎

  9. See (Lu­gar (2005))[http://​​­cle­​​menu/​​key-is­sues/​​nu­clear-weapons/​​is­sues/​​pro­lifer­a­tion/​​fuel-cy­cle/​​sen­ate-dot-gov_NPSur­vey.pdf] ↩︎

  10. See Sand­berg & Bostrom (2008) ↩︎

  11. See Bar­rett et al. (2013) ↩︎

  12. See Apps (2015) ↩︎

  13. Be­cause nu­clear wars are pre­sum­ably very un­likely to cause <1M fatal­ities, I as­sume this is equiv­a­lent to pre­dict­ing the prob­a­bil­ity of any nu­clear war. ↩︎

  14. This es­ti­mate is not di­rectly com­pa­rable to the other es­ti­mates of the prob­a­bil­ity of a nu­clear war be­tween the US and Rus­sia. While the oth­ers the­o­ret­i­cally take into ac­count the prob­a­bil­ity of ac­ci­den­tal nu­clear war and an in­ten­tional nu­clear war, Bar­rett et al.’s 2013 es­ti­mate only con­sider ac­ci­den­tal nu­clear war. I in­clude it in spite of this be­cause I ex­pect most of the prob­a­bil­ity of a US-Rus­sia nu­clear war comes from the po­ten­tial for an ac­ci­den­tal nu­clear ex­change. This makes me think that Bar­rett et al.’s 2013 es­ti­mate is not so differ­ent from an es­ti­mate of the prob­a­bil­ity of any US-Rus­sia nu­clear ex­change. ↩︎

  15. Again, this as­sumes that the ma­jor­ity of the prob­a­bil­ity of a nu­clear at­tack by a state ac­tor caus­ing at least one fatal­ity in Rus­sia comes from a nu­clear ex­change in­volv­ing the US. ↩︎