Book Review: Ages Of Discord

Link post


I re­cently re­viewed Sec­u­lar Cy­cles, which pre­sents a de­mo­graphic-struc­tural the­ory of the growth and de­cline of pre-in­dus­trial civ­i­liza­tions. When land is plen­tiful, pop­u­la­tion grows and the econ­omy pros­pers. When land reaches its car­ry­ing ca­pac­ity and in­come de­clines to sub­sis­tence, the area is at risk of famines, dis­eases, and wars – which kill enough peo­ple that land be­comes plen­tiful again. Dur­ing good times, elites pros­per and act in unity; dur­ing bad times, elites turn on each other in an age of back­stab­bing and civil strife. It seemed pretty rea­son­able, and au­thors Peter Turchin and Sergey Nefe­dov had lots of data to sup­port it.

Ages of Dis­cord is Turchin’s at­tempt to ap­ply the same the­ory to mod­ern Amer­ica. There are many rea­sons to think this shouldn’t work, and the book does a bad job ad­dress­ing them. So I want to start by pre­sent­ing Turchin’s data show­ing such cy­cles ex­ist, so we can at least see why the hy­poth­e­sis might be tempt­ing. Once we’ve seen the data, we can de­cide how turned off we want to be by the the­o­ret­i­cal prob­lems.

The first of Turchin’s two cyclic pat­terns is a long cy­cle of na­tional growth and de­cline. In Sec­u­lar Cy­cles‘ pre-in­dus­trial so­cieties, this pat­tern lasted about 300 years; in Ages of Dis­cord‘s pic­ture of the mod­ern US, it lasts about 150:

This sum­mary figure com­bines many more spe­cific datasets. For ex­am­ple, ar­chae­ol­o­gists fre­quently as­sess the pros­per­ity of a pe­riod by the heights of its skele­tons. Well-nour­ished, happy chil­dren tend to grow taller; a layer with tall skele­tons prob­a­bly rep­re­sents good times dur­ing the rele­vant ar­chae­olog­i­cal pe­riod; one with stunted skele­tons prob­a­bly rep­re­sents famine and stress. What if we ap­plied this to the mod­ern US?

Aver­age US height and life ex­pec­tancy over time. As far as I can tell, the height graph is raw data. The life ex­pec­tancy graph is the raw data minus an as­sumed con­stant pos­i­tive trend – that is, given that tech­nolog­i­cal ad­vance is in­creas­ing life ex­pec­tancy at a lin­ear rate, what are the other fac­tors you see when you sub­tract that out? The ex­act statis­ti­cal logic be buried in Turchin’s source (His­tor­i­cal Statis­tics of the United States, Carter et al 2004), which I don’t have and can’t judge.

This next graph is the me­dian wage di­vided by GDP, a crude mea­sure of in­come equal­ity:

Lower val­ues rep­re­sent more in­equal­ity.

This next graph is me­dian fe­male age at first mar­riage. Turchin draws on re­search sug­gest­ing this tracks so­cial op­ti­mism. In good times, young peo­ple can eas­ily be­come in­de­pen­dent and start sup­port­ing a fam­ily; in bad times, they will want to wait to make sure their lives are sta­ble be­fore set­tling down:

This next graph is Yale tu­ition as a mul­ti­ple of av­er­age man­u­fac­tur­ing worker in­come. To some de­gree this will track in­equal­ity in gen­eral, but Turchin thinks it also mea­sures some­thing like “difficulty of up­ward mo­bil­ity”:

This next graph shows DW-NOMINATE’s “Poli­ti­cal Po­lariza­tion In­dex”, a com­pli­cated met­ric oc­ca­sion­ally used by his­to­ri­ans of poli­tics. It mea­sures the differ­ence in vot­ing pat­terns be­tween the av­er­age Demo­crat in Congress and the av­er­age Repub­li­can in Congress (or for pe­ri­ods be­fore the Democrats and Repub­li­cans, whichever two ma­jor par­ties there were). Dur­ing times of low par­ti­san­ship, con­gres­sional votes will be dom­i­nated by lo­cal or in­di­vi­d­ual fac­tors; dur­ing times of high par­ti­san­ship, it will be dom­i­nated by party iden­ti­fi­ca­tion:

I’ve in­cluded only those graphs which cover the en­tire 1780 – pre­sent pe­riod; the book in­cludes many oth­ers that only cover shorter in­ter­vals (mostly the more re­cent pe­ri­ods when we have bet­ter data). All of them, in­clud­ing the shorter ones not in­cluded here, re­flect the same gen­eral pat­tern. You can see it most eas­ily if you stan­dard­ize all the in­di­ca­tors to the same scale, match the signs so that up always means good and down always means bad, and put them all to­gether:

Note that these aren’t ex­actly the same in­di­ca­tors I fea­tured above; we’ll dis­cuss im­mi­gra­tion later.

The “av­er­age” line on this graph is the one that went into mak­ing the sum­mary graphic above. Turchin be­lieves that af­ter the Amer­i­can Revolu­tion, there was a pe­riod of in­sta­bil­ity last­ing a few decades (eg Shays’ Re­bel­lion, Whiskey Re­bel­lion) but that Amer­ica reached a max­i­mum of unity, pros­per­ity, and equal­ity around 1820. Things grad­u­ally got worse from there, cul­mi­nat­ing in a peak of in­equal­ity, mis­ery, and di­vi­sion around 1900. The re­forms of the Pro­gres­sive Era grad­u­ally made things bet­ter, with an­other unity/​pros­per­ity/​equal­ity max­i­mum around 1960. Since then, an in­creas­ing con­fluence of nega­tive fac­tors (named here as the Rea­gan Era trend re­ver­sal, but Turchin ad­mits it be­gan be­fore Rea­gan) has been mak­ing things worse again.


Along with this “grand cy­cle” of 150 years, Turchin adds a shorter in­sta­bil­ity cy­cle of 40-60 years. This is the same 40-60 year in­sta­bil­ity cy­cle that ap­peared in Sec­u­lar Cy­cles, where Turchin called it “the bi­gen­er­a­tional cy­cle”, or the “fathers and sons cy­cle”.

Timing and in­ten­sity of in­ter­nal war in me­dieval and early mod­ern England, from Turchin and Nefe­dov 2009.

The deriva­tion of this cy­cle, ex­plained on pages 45 – 58 of Ages of Dis­cord, is one of the high­lights of the book. Turchin draws on the kind of mod­els epi­demiol­o­gists use to track pan­demics, think­ing of vi­o­lence as an in­fec­tion and rad­i­cals as plague-bear­ers. You start with an un­ex­posed vuln­er­a­ble pop­u­la­tion. Some rad­i­cal – pa­tient zero – starts call­ing for vi­o­lence. His ideas spread to a cer­tain per­cent of peo­ple he in­ter­acts with, grad­u­ally “in­fect­ing” more and more peo­ple with the “rad­i­cal ideas” virus. But af­ter enough time rad­i­cal­ized, some peo­ple “re­cover” – they be­come ex­hausted with or dis­illu­sioned by con­flict, and be­come pro-co­op­er­a­tion “ac­tive mod­er­ates” who are im­pos­si­ble to re­in­fect (in the epi­demic model, they are “in­oc­u­lated”, but they also have an abil­ity with­out a clear epi­demiolog­i­cal equiv­a­lent to dampen rad­i­cal­ism in peo­ple around them). As the rates of rad­i­cals, ac­tive mod­er­ates, and un­ex­posed dy­nam­i­cally vary, you get a cyclic pat­tern. First ev­ery­one is un­ex­posed. Then rad­i­cal­ism grad­u­ally spreads. Then ac­tive mod­er­a­tion grad­u­ally spreads, un­til it reaches a tip­ping point where it triumphs and rad­i­cal­ism is sup­pressed to a few iso­lated reser­voirs in the pop­u­la­tion. Then the ac­tive mod­er­ates grad­u­ally die off, new un­ex­posed peo­ple are grad­u­ally born, and the cy­cle starts again. Fid­dling with all these var­i­ous pa­ram­e­ters, Turchin is able to get the sys­tem to pro­duce 40-60 year waves of in­sta­bil­ity.

To check this em­piri­cally, Turchin tries to mea­sure the num­ber of “in­sta­bil­ity events” in the US over var­i­ous pe­ri­ods. He very cor­rectly tries to use lists made by oth­ers (since they are harder to bias), but when peo­ple haven’t cat­a­logued ex­actly the kind of in­sta­bil­ity he’s in­ter­ested in over the en­tire 1780 – pre­sent pe­riod, he some­times adds his own in­ter­pre­ta­tion. He ends up sum­ming ri­ots, lynch­ings, ter­ror­ism (in­clud­ing as­sas­si­na­tions), and mass shoot­ings – you can see his defi­ni­tion for each of these start­ing on page 114; the short ver­sion is that all the defi­ni­tions seem rea­son­able but in­evitably in­clude a lot of de­grees of free­dom.

When he adds all this to­gether, here’s what hap­pens:

Poli­ti­cal in­sta­bil­ity /​ vi­o­lent events show three peaks, around 1870, 1920, and 1970.

The 1870 peak in­cludes the Civil War, var­i­ous Civil War as­so­ci­ated vi­o­lence (eg draft ri­ots), and the vi­o­lence around Re­con­struc­tion (in­clud­ing the rise of the Ku Klux Klan and re­lated vi­o­lence to try to con­trol newly eman­ci­pated blacks).

The 1920 peak in­cludes the height of the early US la­bor move­ment. Turchin dis­cusses the Mine War, an “un­de­clared war” from 1920-1921 be­tween bosses and la­bor­ers in Ap­palachian coal coun­try:

Although it started as a la­bor dis­pute, it even­tu­ally turned into the largest armed in­sur­rec­tion in US his­tory, other than the Civil War. Between 10,000 and 15,000 min­ers armed with rifles fought thouas­nds of strike-break­ers and sher­iff’s deputies, called the Lo­gan Defen­ders. The in­sur­rec­tion was ended by the US Army. While such vi­o­lent in­ci­dents were ex­cep­tional, they took place against a back­ground of a gen­eral “class war” that had been in­ten­sify­ing since the vi­o­lent teens. “In 1919 nearly four mil­lion work­ers (21% of the work­force) took dis­rup­tive ac­tion in the face of em­ployer re­luc­tance to rec­og­nize or bar­gain with unions” (Domhoff and Web­ber, 2011:74).

Along with la­bor vi­o­lence, 1920 was also a peak in racial vi­o­lence:

Race-mo­ti­vated ri­ots also peaked around 1920. The two most se­ri­ous such out­breaks were the Red Sum­mer of 1919 (McWhirter 2011) and the Tulsa (Ok­la­homa) Race Riot. The Red Sum­mer in­volved ri­ots in more than 20 cities across the United States and re­sulted in some­thing like 1,000 fatal­ities. The Tulsa riot in 1921, which caused about 300 deaths, took on an as­pect of civil war, in which thou­sands of whites and blacks, armed with firearms, fought in the streets, and most of the Green­wood District, a pros­per­ous black neigh­bor­hood, was de­stroyed.

And ter­ror­ism:

The bomb­ing cam­paign by Ital­ian an­ar­chists (“Galleanists”) cul­mi­nated in the 1920 ex­plo­sion on Wall Street, which caused 38 fatal­ities.

The same prob­lems: la­bor un­rest, racial vi­o­lence, ter­ror­ism – re­peated dur­ing the 1970s spike. In­stead of quot­ing Turchin on this, I want to quote this Sta­tus 451 re­view of Days of Rage, be­cause it blew my mind:

“Peo­ple have com­pletely for­got­ten that in 1972 we had over nine­teen hun­dred do­mes­tic bomb­ings in the United States.” — Max Noel, FBI (ret.)

Re­cently, I had my head torn off by a book: Bryan Bur­rough’s Days of Rage, about the 1970s un­der­ground. It’s the most im­por­tant book I’ve read in a year. So I did a se­ries of run­ning tweet­storms about it, and Clark asked me if he could col­lect them for pos­ter­ity. I’ve ed­ited them slightly for ed­i­to­rial co­her­ence.

Days of Rage is im­por­tant, be­cause this stuff is for­got­ten and it shouldn’t be. The 1970s un­der­ground wasn’t small. It was hun­dreds of peo­ple be­com­ing ur­ban guer­rillas. Bomb­ing build­ings: the Pen­tagon, the Capi­tol, courthouses, restau­rants, cor­po­ra­tions. Rob­bing banks. As­sas­si­nat­ing po­lice. Peo­ple re­ally thought that rev­olu­tion was im­mi­nent, and thought vi­o­lence would bring it about.

One thing that Bur­rough re­turns to in Days of Rage, over and over and over, is how for­got­ten so much of this stuff is. Puerto Ri­can sep­a­ratists bombed NYC like 300 times, kil­led peo­ple, shot up Congress, tried to kill POTUS (Tru­man). No­body re­mem­bers it.

The pas­sage speaks to me be­cause – yeah, no­body re­mem­bers it. This is also how I feel about the 1920 spike in vi­o­lence. I’d heard about the Tulsa race riot, but the Mine War and the bomb­ing of Wall Street and all the other stuff was new to me. This mat­ters be­cause my in­tu­itions be­fore read­ing this book would not have been that there were three gi­ant spikes in vi­o­lence/​in­sta­bil­ity in US his­tory lo­cated fifty years apart. I think the les­son I learn is not to trust my in­tu­itions, and to be a lit­tle more sym­pa­thetic to Turchin’s data.

One more thing: the 1770 spike was ob­vi­ously the Amer­i­can Revolu­tion and all of the ri­ots and com­mu­nal vi­o­lence as­so­ci­ated with it (eg against Tories). Where was the 1820 spike? Turchin ad­mits it didn’t hap­pen. He says that be­cause 1820 was the ab­solute best part of the 150 year grand cy­cle, ev­ery­body was so happy and well-off and pa­tri­otic that the sched­uled in­sta­bil­ity peak just fiz­zled out. Although Turchin doesn’t men­tion it, you could make a similar ar­gu­ment that the 1870 spike was es­pe­cially bad (see: the en­tire frickin’ Civil War) be­cause it hit close to (though not ex­actly at) the worst part of the grand cy­cle. 1920 hit around the mid­dle, and 1970 dur­ing a some­what-good pe­riod, so they fell in be­tween the non­is­sue of 1820 and the dis­aster of 1870.


I haven’t for­got­ten the origi­nal ques­tion – what drives these 150 year cy­cles of rise and de­cline – but I want to stay with the data just a lit­tle longer. Again, these data are re­ally in­ter­est­ing. Either some sort of re­ally in­ter­est­ing the­ory has to be be­hind them – or they’re just low-qual­ity data cherry-picked to make a point. Which are they? Here are a cou­ple of spot-checks to see if the data are any good.

First spot check: can I con­firm Turchin’s data from in­de­pen­dent sources?

Here is a graph of av­er­age US height over time which seems broadly similar to Turchin’s.

Here is a differ­ent mea­sure of US in­come in­equal­ity over time, which again seems broadly similar to Turchin’s. Piketty also pre­sents very similar data, though his story places more em­pha­sis on the World Wars and less on the la­bor move­ment.

– The Columbia Law Re­view mea­sures poli­ti­cal po­lariza­tion over time and gets mostly the same num­bers as Turchin.

I’m go­ing to con­sider this suc­cess­fully checked; Turchin’s data all seem ba­si­cally ac­cu­rate.

Se­cond spot check: do other in­di­ca­tors Turchin didn’t in­clude con­firm the pat­tern he de­tects, or did he just cherry-pick the data se­ries that worked? Spoiler: I wasn’t able to do this one. It was too hard to think of mea­sures that should re­flect gen­eral well-be­ing and that we have 200+ years of un­con­founded data for. But here are my var­i­ous failures:

– The an­nual im­prove­ment in mor­tal­ity rate does not seem to fol­low the cyclic pat­tern. But isn’t this more driven by a few ran­dom fac­tors like smok­ing rates and the logic of tech­nolog­i­cal ad­vance?

Trea­sury bonds maybe kind of fol­low the pat­tern un­til 1980, af­ter which they go crazy.

Divorce rates look kind of iffy, but isn’t that just a bunch of ran­dom fac­tors?

Homi­cide rates, with the gen­eral down­ward trend re­moved, sort of fol­low the pat­tern, ex­cept for the re­cent de­cline?

USD/​GBP ex­change rates don’t show the pat­tern at all, but that could be be­cause of things go­ing on in Bri­tain?

The thing is – re­ally I have no rea­son to ex­pect di­vorce rates, homi­cide rates, ex­change rates etc to track na­tional flour­ish­ing. For one thing, they may just be to­tally un­re­lated. For an­other, even if they were ten­u­ously re­lated, there are all sorts of other ran­dom fac­tors that can af­fect them. The prob­lem is, I would have said this was true for height, age at first mar­riage, and in­come in­equal­ity too, be­fore Turchin gave me con­vinc­ing-sound­ing sto­ries for why it wasn’t. I think my les­son is that I have no idea which in­di­ca­tors should vs. shouldn’t fol­low a sec­u­lar-cyclic pat­tern and so I can’t do this spot check against cherry-pick­ing the way I hoped.

Third spot check: com­mon sense. Here are some things that stood out to me:

– The Civil War is at a low-ish part of the cy­cle, but by no means the low­est.

– The Great De­pres­sion hap­pened at a medium part of the cy­cle, when things should have been quickly get­ting bet­ter.

– Even though there was a lot of new op­ti­mism with Rea­gan, con­tin­u­ing through the Clin­ton years, the cy­cle does not re­flect this at all.

Maybe we can res­cue the first and third prob­lem by com­bin­ing the 150 year cy­cle with the shorter 50 year cy­cle. The Civil War was de­ter­mined by the 50-year cy­cle hav­ing its oc­ca­sional burst of vi­o­lence at the same time the 150-year cy­cle was at a low-ish point. Peo­ple have good mem­o­ries of Rea­gan be­cause the chaos of the 1970 vi­o­lence burst had ended.

As for the sec­ond, Turchin is aware of the prob­lem. He writes:

There is a widely held be­lief among economists and other so­cial sci­en­tists that the 1930s were the “defin­ing mo­ment” in the de­vel­op­ment of the Amer­i­can poli­tico-eco­nomic sys­tem (Bordo et al 1998). When we look at the ma­jor struc­tural-de­mo­graphic vari­ables, how­ever, the decade of the 1930s does not seem to be a turn­ing point. Struc­tural-de­mo­graphic trends that were es­tab­lished dur­ing the Pro­gres­sive Era con­tinues through the 1930s, al­though some of them ac­cel­er­ated.

Most no­tably, all the well-be­ing vari­ables that went through trend re­ver­sals be­fore the Great De­pres­sion – be­tween 1900 and 1920. From roughly 1910 and to 1960 they all in­creased roughly mono­ton­i­cally, with only one or two minor fluc­tu­a­tions around the up­ward trend. The dy­nam­ics of real wages also do not ex­hibit a break­ing point in the 1930s, al­though there was a minor ac­cel­er­a­tion af­ter 1932.

By com­par­i­son, he plays up the con­ve­niently-timed (and hith­erto un­known to me) de­pres­sion of the mid-1890s. Quot­ing Turchin quot­ing McCormick:

No de­pres­sion had ever been as deep and tragic as the one that lasted from 1893 to 1897. Millions suffered un­em­ploy­ment, es­pe­cially dur­ing the win­ters of 1893-4 and 1894-5, and thou­sands of ‘tramps’ wan­dered the coun­tryside in search of food […]

De­spite real hard­ship re­sult­ing form mas­sive un­em­ploy­ment, well-be­ing in­di­ca­tors sug­gest that the hu­man cost of the Great De­pres­sion of the 1930s did not match that of the “First Great De­pres­sion” of the 1890s (see also Grant 1983:3-11 for a gen­eral dis­cus­sion of the sever­ity of the 1890s de­pres­sion. Fur­ther­more, while the 1930s are re­mem­bered as a pe­riod of vi­o­lent la­bor un­rest, the in­ten­sity of class strug­gle was ac­tu­ally lower than dur­ing the 1890s de­pres­sion. Ac­cord­ing to the US Poli­ti­cal Violence Database (Turchin et al. 2012) there were 32 lethal la­bor dis­putes dur­ing the 1890s that col­lec­tively caused 140 deaths, com­pared with 20 such dis­putes in the 1930s with the to­tal of 55 deaths. Fur­ther­more, the last lethal strike in US la­bor his­tory was in 1937…in other words, the 1930s was ac­tu­ally the last uptick of vi­o­lent class strug­gle in the US, su­per­im­posed on an over­all de­clin­ing trend.

The 1930s De­pres­sion is prob­a­bly re­mem­bered (or rather mis­re­mem­bered) as the worst eco­nomic slump in US his­tory, sim­ply be­cause it was the last of the great de­pres­sions of the post-Civil War era.

Fourth spot check: Did I ran­domly no­tice any egre­gious er­rors while read­ing the book?

On page 70, Turchin dis­cusses “the great cholera epi­demic of 1849, which car­ried away up to 10% of the Amer­i­can pop­u­la­tion”. This seemed un­be­liev­ably high to me. I checked the source he cited, Kohl’s “En­cy­clo­pe­dia Of Plague And Pestilence”, which did give that num­ber. But ev­ery other source I checked agreed that the epi­demic “only” kil­led be­tween 0.3% – 1% of the US pop­u­la­tion (it did hit 10% in a few es­pe­cially un­lucky cities like St. Louis). I can­not fault Turchin’s schol­ar­ship in the sense of cor­rectly re­peat­ing some­thing writ­ten in an en­cy­clo­pe­dia, but un­less I’m miss­ing some­thing I do fault his com­mon sense.

Also, on page 234, Turchin in­ter­prets the per­cent of med­i­cal school grad­u­ates who get a res­i­dency as “the gap be­tween the de­mand and sup­ply of MD po­si­tions”, which he ties into a wider ar­gu­ment about elite over­pro­duc­tion. But I think this shows a limited un­der­stand­ing of how the med­i­cal sys­tem works. There is cur­rently a se­vere un­der­sup­ply of doc­tors – try get­ting an ap­point­ment with a spe­cial­ist who takes in­surance in a rea­son­able amount of time if you don’t be­lieve me. Res­i­den­cies aren’t limited by or­ganic de­mand. They’re limited be­cause the gov­ern­ment places so many re­stric­tions on them that hos­pi­tals don’t spon­sor them with­out gov­ern­ment fund­ing, and the gov­ern­ment is too stingy to fund more of them. None of this has any­thing to do with elite over­pro­duc­tion.

Th­ese are just two small er­rors in a long book. But they’re two er­rors in medicine, the field I know some­thing about. This makes me worry about Gell-Mann Am­ne­sia: if I no­tice er­rors in my own field, how many er­rors must there be in other fields that I just didn’t catch?

My over­all con­clu­sion from the spot-checks is that the data as pre­sented are ba­si­cally ac­cu­rate, but that ev­ery­thing else is so de­pen­dent on liti­gat­ing which things are vs. aren’t in ac­cor­dance with the the­ory that I ba­si­cally give up.


Okay. We’ve gone through the data sup­port­ing the grand cy­cle. We’ve gone through the data and the­ory for the 40-60 year in­sta­bil­ity cy­cle. We’ve gone through the rea­sons to trust vs. dis­trust the data. Time to go back to the ques­tion we started with: why should the grand cy­cle, origi­nally de­rived from the Malthu­sian prin­ci­ples that gov­ern pre-in­dus­trial so­cieties, hold in the mod­ern US? Food and land are no longer limit­ing re­sources; famines, dis­ease, and wars no longer sub­stan­tially de­crease pop­u­la­tion. Al­most ev­ery fac­tor that drives the origi­nal sec­u­lar cy­cle is miss­ing; why even con­sider the pos­si­bil­ity that it might still ap­ply?

I’ve put this off be­cause, even though this is the ob­vi­ous ques­tion Ages of Dis­cord faces from page one, I found it hard to get a sin­gle clear an­swer.

Some­times, Turchin talks about the sup­ply vs. de­mand of la­bor. In times when the sup­ply of la­bor out­paces de­mand, wages go down, in­equal­ity in­creases, elites frag­ment, and the coun­try gets worse, mimick­ing the “land is at car­ry­ing ca­pac­ity” stage of the Malthu­sian cy­cle. In times when de­mand for la­bor ex­ceeds sup­ply, wages go up, in­equal­ity de­creases, elites unite, and the coun­try gets bet­ter. The gov­ern­ment is con­trol­led by plu­to­crats, who always want wages to be low. So they im­ple­ment poli­cies that in­crease the sup­ply of la­bor, es­pe­cially loose im­mi­gra­tion laws. But their ac­tions cause in­equal­ity to in­crease and ev­ery­one to be­come mis­er­able. Or­di­nary peo­ple or­ga­nize re­sis­tance: pop­ulist move­ments, so­cial­ist cadres, la­bor unions. The sys­tem teeters on the edge of vi­o­lence, rev­olu­tion, and to­tal dis­in­te­gra­tion. Since the elites don’t want those things, they take a step back, re­al­ize they’re kil­ling the goose that lays the golden egg, and de­cide to loosen their grip on the neck of the pop­u­lace. The gov­ern­ment be­comes mod­er­ately pro-la­bor and pro­gres­sive for a while, and tight­ens im­mi­gra­tion laws. The over­sup­ply of la­bor de­creases, wages go up, in­equal­ity goes down, and ev­ery­one is happy. After ev­ery­one has been happy for a while, the pop­ulists/​so­cial­ists/​unions lose rele­vance and drift apart. A new gen­er­a­tion of elites who have never felt threat­ened come to power, and they think to them­selves “What if we used our con­trol of the gov­ern­ment to squeeze la­bor harder?” Thus the cy­cle be­gins again.

But at other times, Turchin talks more about “elite over­pro­duc­tion”. When there are rel­a­tively few elites, they can co­op­er­ate for their com­mon good. Bi­par­ti­san­ship is high, ev­ery­one is unified be­hind a sys­tem per­ceived as wise and benev­olent, and we get a his­tor­i­cal pe­riod like the 1820s US golden age that his­to­ri­ans call The Era Of Good Feel­ings. But as the num­ber of elites out­strips the num­ber of high-sta­tus po­si­tions, com­pe­ti­tion heats up. Elites re­al­ize they can get a leg up in an in­creas­ingly difficult rat race by back­stab­bing against each other and the coun­try. Govern­ment and cul­ture en­ter a defect-defect era of hy­per­par­ti­san­ship, where ev­ery­one burns the com­mons of pro­duc­tive norms and in­sti­tu­tions in or­der to get ahead. Even­tu­ally…some pro­cess re­verses this or some­thing?…and then the cy­cle starts again.

At still other times, Turchin seems to re­treat to a sort of math­e­mat­i­cal for­mal­ism. He con­structs an ex­tremely hokey-look­ing dy­namic feed­back model, based on ideas like “as­sume that the level of dis­con­tent among or­di­nary peo­ple equals the ur­ban­iza­tion rate x the age struc­ture x the in­verse of their wages rel­a­tive to the elite” or “let us define the fis­cal dis­tress in­dex as debt ÷ GDP x the level of dis­trust in state in­sti­tu­tions”. Then he puts these all to­gether into a model that calcu­lates how the the level of dis­con­tent af­fects and is af­fected by the level of state fis­cal dis­tress and a few dozen other vari­ables. On the one hand, this is re­ally cool, and watch­ing it in ac­tion gives you the same kind of feel­ing Sel­don must have had in­vent­ing psy­chohis­tory. On the other, it seems re­ally made-up. Turchin ad­mits that dy­namic feed­back sys­tems are in­fa­mous for go­ing com­pletely hay­wire if they are even a tiny bit skew to re­al­ity, but as­sures us that he un­der­stands the cut­ting-edge of the field and how to make them not to do that. I don’t know enough to judge whether he’s right or wrong, but my pri­ors are on “ex­tremely, al­most un­fath­omably wrong”. Still, at times he re­minds us that the shifts of dy­namic feed­back sys­tems can be at­tributed only to the sys­tem in its en­tirety, and that try­ing to tell sto­ries about or point to spe­cific fac­tors in­volved in any par­tic­u­lar shift is an ap­prox­i­ma­tion at best.

All of these three sto­ries run into prob­lems al­most im­me­di­ately.

First, the sup­ply of la­bor story fo­cuses pretty heav­ily on im­mi­gra­tion. Turchin puts a lot of work into show­ing that im­mi­gra­tion fol­lows the sec­u­lar cy­cle pat­terns; it is high­est at the worst part of the cy­cle, and low­est at the best parts:

In this model, im­mi­gra­tion is a tool of the plu­toc­racy. High sup­ply of la­bor (rel­a­tive to de­mand) drives down wages, in­creases in­equal­ity, and low­ers work­ers’ bar­gain­ing power. If the la­bor sup­ply is poorly or­ga­nized, comes from places that don’t un­der­stand the con­cept of “union”, don’t know their rights, and have racial and lin­guis­tic bar­ri­ers pre­vent­ing them from co­op­er­at­ing with the rest of the work­ing class, well, even bet­ter. Thus, pe­ri­ods when the plu­toc­racy is suc­cess­fully squeez­ing the work­ing class are marked by high im­mi­gra­tion. Pe­ri­ods when the plu­toc­racy fears the work­ing class and feels com­pel­led to be nice to them are marked by low im­mi­gra­tion.

This po­si­tion makes some sense and is loosely sup­ported by the long-term data above. But isn’t this one of the most-stud­ied top­ics in the his­tory of eco­nomics? Hasn’t it been proven al­most be­yond doubt that im­mi­grants don’t steal jobs from Amer­i­can work­ers, and that since they con­sume prod­ucts them­selves (and thus in­crease the de­mand for la­bor) they don’t af­fect the sup­ply/​de­mand bal­ance that sets wages?

It ap­pears I might just be to­tally mis­cal­ibrated on this topic. I checked the IGM Eco­nomic Ex­perts Panel. Although most of the ex­pert economists sur­veyed be­lieved im­mi­gra­tion was a net good for Amer­ica, they did say (50% agree to only 9% dis­agree) that “un­less they were com­pen­sated by oth­ers, many low-skil­led Amer­i­can work­ers would be sub­stan­tially worse off if a larger num­ber of low-skil­led for­eign work­ers were legally al­lowed to en­ter the US each year”. I’m hav­ing trou­ble see­ing the differ­ence be­tween this state­ment (which economists seem very con­vinced is true) and “you should worry about im­mi­grants steal­ing your job” (which ev­ery­one seems very con­vinced is false). It might be some­thing like – im­mi­gra­tion gen­er­ally makes “the econ­omy bet­ter”, but there’s no guaran­tee that these gains are evently dis­tributed, and so it can be bad for low-skil­led work­ers in par­tic­u­lar? I don’t know, this would still rep­re­sent a pretty big up­date, but given that I was told all top economists think one thing, and now I have a sur­vey of all top economists say­ing the other, I guess big up­dates are un­avoid­able. In­ter­ested in hear­ing from some­one who knows more about this.

Even if it’s true that im­mi­gra­tion can hurt low-skil­led work­ers, Turchin’s po­si­tion – which is that in­creased im­mi­gra­tion is re­spon­si­ble for a very large por­tion of post-1973 wage stag­na­tion and the re­cent trend to­ward ris­ing in­equal­ity – sounds shock­ing to cur­rent poli­ti­cal sen­si­bil­ities. But all Turchin has to say is:

An im­bal­ance be­tween la­bor sup­ply and de­mand clearly played an im­por­tant role in driv­ing real wages down af­ter 1978. As Har­vard economist Ge­orge J. Bor­jas re­cently wrote, “The best em­piri­cal re­search that tries to ex­am­ine what has ac­tu­ally hap­pened in the US la­bor mar­ket al­igns well with eco­nomic the­ory: An in­crease in the num­ber of work­ers leads to lower wages.”

My im­pres­sion was that Bor­jas was an in­creas­ingly iso­lated con­trar­ian voice, so once again, I just don’t know what to do here.

Se­cond, the plu­to­cratic op­pres­sion story re­lies pretty heav­ily on the idea that in­equal­ity is a unique bad. This fits the zeit­geist pretty well, but it’s a lit­tle con­fus­ing. Why should com­mon­ers care about their wages rel­a­tive to elites, as op­posed to their ab­solute wages? Although me­dian-wage-rel­a­tive-to-GDP has gone down over the past few decades, ab­solute me­dian wage has gone up – just a lit­tle, slowly enough that it’s rightly con­sid­ered a prob­lem – but it has gone up. Since mod­ern wages are well above 1950s wages, in way sense should mod­ern peo­ple feel like they are eco­nom­i­cally bad off in a way 1950s peo­ple didn’t? This isn’t a prob­lem for Turchin’s the­ory so much as a gen­eral mys­tery, but it’s a gen­eral mys­tery I care about a lot. One an­swer is that the cost dis­ease is fueled by a Bau­mol effect pegged to per cap­i­tal in­come (see part 3 here), and this is a way that in­creas­ing elite wealth can ab­solutely (not rel­a­tively) im­miser­ate the lower classes.

Like­wise, what about The Spirit Level Delu­sion and other re­sources show­ing that, across coun­tries, in­equal­ity is not par­tic­u­larly cor­re­lated with so­cial bads? Does this challenge Turchin’s Amer­ica-cen­tric find­ings that ev­ery­thing gets worse along with in­equal­ity lev­els?

Third, the plu­to­cratic op­pres­sion story meshes poorly with the elite over­pro­duc­tion story. In elite over­pro­duc­tion, united elites are a sign of good times to come; di­vided elites means dys­func­tional gov­ern­ment and po­ten­tial vi­o­lence. But as Pseu­do­eras­mus points out, united elites are of­ten united against the com­mon­ers, and we should ex­pect in­equal­ity to be high­est at times when the elites are able to work to­gether to fight for a larger share of the pie. But I think this is the op­po­site of Turchin’s story, where elites unite only to make con­ces­sions, and elite unity equals pop­u­lar pros­per­ity.

Fourth, ev­ery­thing about the elite over­pro­duc­tion story con­fuses me. Who are “elites”? This cat­e­gory made sense in Sec­u­lar Cy­cles, which dis­cussed agrar­ian so­cieties with a dis­tinct ti­tled no­bil­ity. But Turchin wants to define US elites in terms of wealth, which fol­lows a con­tin­u­ous dis­tri­bu­tion. And if you’re defin­ing elites by wealth, it doesn’t make sense to talk about “not enough high-sta­tus po­si­tions for all elites”; if you’re elite (by virtue of your great wealth), by defi­ni­tion you already have what you need to main­tain your elite sta­tus. Turchin seems aware of this is­sue, and some­times talks about “elite as­pirants” – some kind of up­per class who ex­pect to be wealthy, but might or might not get that as­pira­tion fulfilled. But then un­der­stand­ing elite over­pro­duc­tion hinges on what makes one non-rich-per­son per­son a com­moner vs. an­other non-rich-per­son an “elite as­pirant”, and I don’t re­mem­ber any clear dis­cus­sion of this in the book.

Fifth, what drives elite over­pro­duc­tion? Why do elites (as a per­cent of the pop­u­la­tion) in­crease dur­ing some pe­ri­ods and de­crease dur­ing oth­ers? Why should this be a cy­cle rather than a ran­dom walk?

My guess is that Ages of Dis­cord con­tains an­swers to some of these ques­tions and I just missed them. But I missed them af­ter read­ing the book pretty closely to try to find them, and I didn’t feel like there were any similar holes in Sec­u­lar Cy­cles. As a re­sult, al­though the book had some fas­ci­nat­ing data, I felt like it lacked a clear and lu­cid the­sis about ex­actly what was go­ing on.


Ac­cept­ing the data as ba­si­cally right, do we have to try to wring some sense out of the the­ory?

The data cover a cy­cle and a half. That means we only sort of barely get to see the cy­cle “re­peat”. The con­clu­sion that it is a cy­cle and not some dis­con­nected trends is based only on the sin­gle co­in­ci­dence that it was 70ish years from the first turn­ing point (1820) to the sec­ond (1890), and also 70ish years from the sec­ond to the third (1960).

A par­si­mo­nious ex­pla­na­tion would be “for some rea­son things were go­ing un­usu­ally well around 1820, un­usu­ally badly around 1890, and un­usu­ally well around 1960 again.” This is ac­tu­ally re­ally in­ter­est­ing – I didn’t know it was true be­fore read­ing this book, and it changes my con­cep­tion of Amer­i­can his­tory a lot. But it’s a lot less in­ter­est­ing than the dis­cov­ery of a sec­u­lar cy­cle.

I think the par­si­mo­nious ex­pla­na­tion is close to what Thomas Piketty ar­gued in his Cap­i­tal In The Twenty-First Cen­tury. Inequal­ity was ris­ing un­til the World Wars, be­cause that’s what in­equal­ity nat­u­rally does given rea­son­able as­sump­tions about growth rates. Then the De­pres­sion and World Wars wiped out a lot of ex­ist­ing money and power struc­tures and made things equal again for a lit­tle while. Then in­equal­ity started ris­ing again, be­cause that’s what in­equal­ity nat­u­rally does given rea­son­able as­sump­tions about growth rates. Add in a pinch of The Spirit Level – in­equal­ity is a mys­te­ri­ous magic poi­son that some­how makes ev­ery­thing else worse – and there’s not much left to be ex­plained.

(some ex­cep­tions: why was in­equal­ity de­creas­ing un­til 1820? Does in­equal­ity re­ally drive poli­ti­cal po­lariza­tion? When im­mi­gra­tion cor­re­sponds to pe­ri­ods of high in­equal­ity, is the im­mi­gra­tion caus­ing the in­equal­ity? And what about the 50 year cy­cle of vi­o­lence? That’s an­other co­in­ci­dence we didn’t in­clude in the co­in­ci­dence list!)

So what can we get from Ages of Dis­cord that we can’t get from Piketty?

First, the con­cept of “elite over­pro­duc­tion” is one that worms its way into your head. It’s the sort of thing that was con­stantly in the back­ground of In­creas­ingly Com­pet­i­tive Col­lege Ad­mis­sions: Much More Than You Wanted To Know. It’s the sort of thing you think about when a mil­lion fresh-faced col­lege grad­u­ates want to be­come Jour­nal­ists and Shape The Con­ver­sa­tion and Fight For Jus­tice and re­al­is­ti­cally just end up get­ting ground up and spit out by click­bait web­sites. Ages of Dis­cord didn’t do a great job break­ing down its ex­act dy­nam­ics, but I’m grate­ful for its work bring­ing it from a sort of shared un­con­scious as­sump­tion into the light where we can talk about it.

Se­cond, the idea of a deep link be­tween var­i­ous in­di­ca­tors of good­ness and bad­ness – like wages and par­ti­san po­lariza­tion – is an im­por­tant one. It forces me to reeval­u­ate things I had con­sid­ered set­tled, like that im­mi­gra­tion doesn’t worsen in­equal­ity, or that in­equal­ity is not a mag­i­cal curse that poi­sons ev­ery­thing.

Third, his­to­ri­ans have to choose what events to fo­cus on. Nor­mal his­to­ri­ans usu­ally fo­cus on the same nor­mal events. Unusual his­to­ri­ans some­times fo­cus on ne­glected events that sup­port their un­usual the­ses, so read­ing some­one like Turchin is a good way to learn parts of his­tory you’d never en­counter oth­er­wise. Some of these I was able to men­tion above – like the Mine War of 1920 or the cholera epi­demic of 1849; I might make an­other post for some of the oth­ers.

Fourth, it tries to link events most peo­ple would con­sider sep­a­rate – wage stag­na­tion since 1973, the Great Stag­na­tion in tech­nol­ogy, the de­cline of Peter Thiel’s “definite op­ti­mism”, the rise of par­ti­san po­lariza­tion. I’m not sure ex­actly how it links them or what it has to stay about the link, but link them it does.

But the most im­por­tant thing about this book is that Turchin claims to be able to pre­dict the fu­ture. The book (writ­ten just be­fore Trump was elected in 2016) ends by say­ing that “we live in times of in­ten­sify­ing struc­tural-de­mo­graphic pres­sures for in­sta­bil­ity”. The next bi­gen­er­a­tional burst of vi­o­lence is sched­uled for about 2020 (re­al­is­ti­cally +/​- a few years). It’s at a low point in the grand cy­cle, so it should be a doozy.

What about be­yond that? It’s un­clear ex­actly where he thinks we are right now in the grand cy­cle. If the cur­rent cy­cle lasts ex­actly as long as the last one, we would ex­pect it to bot­tom out in 2030, but Turchin never claims ev­ery cy­cle is ex­actly as long. A few of his graphs sug­gest a hint of cur­va­ture, sug­gest­ing we might cur­rently be in the worst of it. The so­cial­ists seem to have got­ten their act to­gether and be­come an im­por­tant poli­ti­cal force, which the the­ory pre­dicts is a nec­es­sary pre­cur­sor to change.

I think we can count the book as hav­ing made cor­rect pre­dic­tions if vi­o­lence spikes in the very near fu­ture (are the cur­rent num­ber of mass shoot­ings enough to satisfy this re­quire­ment? I would have to see it graphed us­ing the same mea­sure­ments as past spikes), and if some­time in the next decade or so things start look­ing like there’s a ray of light at the end of the tun­nel.

I am pretty in­ter­ested in find­ing other ways to test Turchin’s the­o­ries. I’m go­ing to ask some of my math ge­nius friends to see if the dy­namic feed­back mod­els check out; if any­one wants to help, let me know how I can help you (if money is an is­sue, I can send you a copy of the book, and I will definitely pub­lish any­thing you find on this blog). If any­one has any other ideas for to in­di­ca­tors that should be cor­re­lated with the sec­u­lar cy­cle, and ideas about how to find them, I’m in­tereted in that too. And if any­one thinks they can ex­plain the elite over­pro­duc­tion is­sue, please en­lighten me.

I ended my re­view of Sec­u­lar Cy­cles by say­ing:

One thing that strikes me about [Turchin]’s cy­cles is the ide­olog­i­cal com­po­nent. They de­scribe how, dur­ing a growth phase, ev­ery­one is op­ti­mistic and pa­tri­otic, se­cure in the knowl­edge that there is enough for ev­ery­body. Dur­ing the stagfla­tion phase, in­equal­ity in­creases, but con­cern about in­equal­ity in­creases even more, zero-sum think­ing pre­dom­i­nates, and so­cial trust craters (both be­cause peo­ple are ac­tu­ally defect­ing, and be­cause it’s in lots of peo­ple’s in­ter­est to play up the de­gree to which peo­ple are defect­ing). By the crisis phase, par­ti­san­ship is much stronger than pa­tri­o­tism and rad­i­cals are talk­ing openly about how vi­o­lence is eth­i­cally obli­ga­tory.

And then, even­tu­ally, things get bet­ter. There is a new Au­gus­tan Age of virtue and the reestab­lish­ment of all good things. This is a re­ally in­ter­est­ing claim. Western philos­o­phy tends to think in terms of trends, not cy­cles. We see ev­ery­thing go­ing on around us, and we think this is some end­less trend to­wards more par­ti­san­ship, more in­equal­ity, more ha­tred, and more state dys­func­tion. But Sec­u­lar Cy­cles offers a nar­ra­tive where end­less trends can end, and things can get bet­ter af­ter all.

This is still the hope, I guess. I don’t have a lot of faith in hu­man effort to re­store nice­ness, com­mu­nity, and civ­i­liza­tion. All I can do is pray the Vast Form­less Things ac­com­plish it for us with­out ask­ing us first.