Data Essays

These are short data essays drawn from my research in progress. Most run the data behind a working paper to answer a question the paper itself does not take up.

Printed weapons know-how traveled only where readers already had the capital

The worry about AI and weapons is that a model explaining how to build one puts it in anyone’s hands. Europe ran a version of that experiment with print. The Universal Short Title Catalogue holds 2,182 editions of military manuals printed before 1650, and after 1550 about nine in ten were in the vernacular. The cannon books barely traveled. Biringuccio’s manual of gunpowder and cannon founding took 16 years to reach a second language and Tartaglia’s ballistics was never translated before 1650. De Gheyn’s musket drill of 1607 was in four languages within a year. My reading is that print moved know-how fast only where readers already had the foundry or the men and the muskets. Whether any of it caused harm has not been measured.

Read the whole thread

1/ The AI-safety worry is that a model explaining how to build a weapon puts it in anyone’s hands. Europe ran this experiment with print. Gunpowder, cannon founding, ballistics, and drill went into print in the 1500s. Whether the know-how spread depended on what else you needed.

2/ Europe printed this harmful technology openly, some of it under state privilege. The Universal Short Title Catalogue holds 2,182 editions of military manuals before 1650. After 1550 about nine in ten were vernacular, against about six in ten for all print.

3/ The authors knew it. In 1537 Tartaglia wrote that perfecting gunnery was “a reproachful, vituperative and cruel thing” and that he had “shredded and burned” his notes. Then, seeing “the wolf wishing to join our flock” (in context, the Ottomans), he published anyway.

4/ Yet the cannon books barely traveled. Biringuccio’s manual of gunpowder and cannon founding took 16 years to reach a second language. Tartaglia’s ballistics was never translated before 1650. Valturio’s war machines waited 83 years. Nine, five, and eight editions in all.

5/ My read is that the book was never the bottleneck. A cannon needed a foundry, a powder works, costly bronze, and a master founder trained by doing. Guns had been in Europe since the 1320s, more than a century before the press. Print supplied the one input that was not scarce.

6/ Compare drill. In 1607 De Gheyn printed Maurice of Nassau’s musket and pike drill. It was out in Dutch, English, French, and German by 1608, 25 editions in all. Armies already had the men and the muskets. The missing piece was the know-how and print moved it fast.

7/ Cheap know-how moved fastest of all. Alessio’s book of secrets (1555), mostly remedies, cosmetics, and household recipes, ran to 154 editions in seven languages, most of them pocket-sized. Did any of it cause harm? Nobody has measured that.

8/ Charles Jones (JEP 2026) argues AI’s payoff is held back by weak links, the tasks that stay slow or physical. I’d add that harm has weak links too and capital is one. It stops actors who lack it. It does little against those who already have it, as the drill manual shows.

9/ Caveats. These are surviving editions in one catalog, so lost printings are missing. Translation speed measures demand, and harm goes unmeasured. De Gheyn’s buyers were officers and states, so the drill case says nothing about lone actors. The capital argument is my reading.

10/ Method: every edition of each book to 1650 in the Universal Short Title Catalogue, checked by hand, dated by first appearance in each language. Tartaglia quoted in Valleriani’s translation. Which other dangerous book should I check? /end

Method. Every edition of each book to 1650 in the Universal Short Title Catalogue, checked by hand and dated by first appearance in each language. Tartaglia is quoted from Valleriani’s English translation (Edition Open Sources, 2013). The weak links argument is from C. I. Jones, “AI and Our Economic Future,” Journal of Economic Perspectives 40(3), 2026. Images are public domain via Wikimedia Commons. The write up, the editions and the code are here.

Will AI standardize the way we write?

Arthur Spirling asked whether AI-assisted writing will standardize language further than the printing press did. The printing press is a version of that experiment we have already run. In the 1540s three of every four English title pages spelled it “churche.” By the 1590s nine in ten wrote “church.” Printers reached one spelling by two roads, a dominant printing center like London or a shared reference like the classical texts behind Latin. Where they had neither, as in Germany’s many printing towns, variety lasted a century or more. My guess is that AI will standardize faster because it has both roads at once. The catalog’s transcriptions limit what the curves can show.

Read the whole thread

1/ We have run a version of the AI experiment already, with the printing press. In the 1540s, 3 of 4 English title pages spelled it “churche.” By the 1590s, 9 of 10 wrote “church.” We use 788,295 European titles to show: where print standardized fast, where it never did, and what that predicts for AI.

2/ The prediction first. AI will standardize writing faster than print did. Print reached one spelling by two roads, a dominant center or a shared reference text. Where it had neither, variety lasted a century or more. AI has both roads at once. The evidence follows…

3/ In 1490 Caxton asked what a man should write: “egges or eyren”? A century on, print had settled questions like it. Across 17 words in 52,000 English titles, the leading spelling’s share rose from 66% in the 1550s to 93% in the 1590s. The first English-only dictionary: 1604.

4/ Why so fast? My hypothesis: one town, then one guild. London printed 90% of English titles in 1540–99. In 1557 a royal charter put English printing under one London guild, the Stationers’ Company. Convergence starts in the 1570s. German needed 25 cities to print 80% of its titles.

5/ German converged only after 1600 and its towns kept house styles. In 1570–99 Wittenberg printed “deudsch” (91%) and Nuremberg “teutsch” (81%). By the 1630s each big town was 99.7% consistent at home, and they still disagreed: Vienna wrote “Zeittung,” Hamburg “Zeitung.”

6/ Dutch went the other way. Antwerp printed 57% of Dutch titles before 1540. After the Revolt split the market, Amsterdam led (37% after 1600). In 1620–49 Antwerp still wrote “coninck” (86%). Amsterdam was torn: coninck 45%, koninck 30%, coningh 21%. A new center, a new fight.

7/ Latin breaks the rule. The most scattered market of all — 31 cities for 80% — converged first. Classical “ecclesiae” went from 11% of uses in 1501–09 to 94% by the 1560s, over medieval “ecclesie.” Latin printers shared no town. They shared a reference: the classical texts.

8/ So print had two roads to one spelling: a dominant center (London) or a shared reference (classical Latin). Where neither held — Germany’s 200+ towns, a Dutch market split between two states — variety lasted to 1650. The leading spelling’s share, four languages:

9/ Back to AI. A handful of models trained on much the same text: a London and a classical canon at once. And AI writes the words, so the form is set at composition. What would prove me wrong? AI text that varies by model the way German varied by town. My interpretation/guess, not a result.

10/ Caveats. These are catalogue transcriptions: long s is in 0 of 788,295 titles, u/v is modernized, English USTC records change style at 1600. We merge u/v and i/j and read trends inside one regime. Five languages is a small sample. Most pre-1600 German variety sat inside each town.

11/ Method: 54 words, four languages. We hand-checked which spellings are one word and took the leading spelling’s share per decade. The groupings are our call. French is left to @jrhall97 and Alex Taylor, whose paper on the 1539 ordinance does it right. Send me comments or questions. /end

Method. Spelling families for 54 common title words in four languages from the Universal Short Title Catalogue, hand-checked for which spellings are one word, with the leading spelling’s share computed by decade. The groupings are our call. French is left to Taylor and Hall, whose paper “The King’s French” (2026) uses a border design around the 1539 Ordinance of Villers-Cotterêts. The word lists, tables and code for every number are here.

The word for the state turned political before the state could tax

In 1500 the state of something was a condition you were in, such as matrimony, grace or old age. By 1640 it was a kingdom. That crossover is visible on 52,000 English title pages, while tax revenue per head across five European powers stayed nearly flat from 1500 to 1700. The vocabulary arrived first. The first eight English titles to call a place a state, between 1570 and 1579, all named foreign places, so the word entered English as news of other people’s wars. Spain never crosses and was the superpower, so this is a poor gauge of state capacity and a good record of when an abstraction became available.

Read the whole thread

1/ State capacity matters. So does language. The early-modern print record shows both changing, through one word. In 1500 “the state of ___” was your condition: matrimony, grace, old age. By 1640 it was the kingdom. 52,000 English title pages, decade by decade.

2/ What the word came to name — a treasury, an army, a debt that outlives the king — was barely moving. Tax revenue per head was nearly flat from 1500 to 1700. The word crossed over in those centuries. You can’t fund a perpetual debt until there’s a perpetual entity to owe it.

3/ Seven languages, one blank after “state of”. Every vernacular flips to polities — French by 1550, English by 1570 — and Latin never does. Yet it’s a poor gauge of capacity: Spain, the superpower, never crosses; England crosses before it can tax; the Dutch get both at once.

4/ Here’s the part nobody had shown. The first eight times an English title called a place a “state” — 1570 to 1579 — it was Germany, Holland, France, Flanders, Venice. Every one foreign. England isn’t named until 1585. The word arrived as news of other people’s wars.

5/ Where did the thing come from? From the state printing under its own name, in the vernacular, before anyone else. Royal edicts were 95–100% vernacular from 1500, when the book market was a third to two-thirds Latin. Latin’s own “ratio status” doesn’t appear until after 1600.

6/ Language as statecraft. The French king’s edicts were 95–100% French from 1500. Villers-Cotterêts (1539) made French the language of law: legal print went from 12% French in the 1530s to 66% by the 1590s, the market 35% to 66%. Taylor & @jrhall97 (2026) show it caused that.

7/ Two cases break “no word, no state”. Germany: 10,048 ordinances in German, 7 uses of “Staat” — the apparatus without the word. The Dutch: last to the word and most state-like with it, because their governing body was the States. Their 1637 Bible is the Statenvertaling.

8/ And Italy, which rose first and then “fell”? After 1600 the word stops taking a complement and starts standing alone — ragion di stato, governo di stato — while devotional print about the state of souls surges. The word moved on.

9/ Method & caveats. Regex on title pages and hand-coded word lists — no AI, no embeddings; all linked below. Thin early cells: 18 English uses before 1550. Titles aren’t prose. Reprints are collapsed. Italy leans on Milan’s edicts; without them its peak falls from 75% to 51%.

10/ The lineage: Skinner dated the state from a dozen texts; here it is in 788,000. Koselleck puts German “Staat” in the 17th c.; 7 hits agree. Anderson said print made nations — the state printed first. Kantorowicz’s state that outlives the king is the one that can owe. /end

Method. Regular expressions over title pages in the Universal Short Title Catalogue, with hand coded word lists, across seven languages. No embeddings and no AI classification in this one. The fiscal figure is from Johnson and Koyama (2017), drawing on Karaman and Pamuk (2013). The Villers-Cotterêts result is from Taylor and Hall, “The King’s French” (2026). The full write up, the word lists and the figures are here.

Guild cities were twice the size of other cities four centuries before their guilds were recorded

Craft guilds first appear in the record in cities that had already been about twice the size of other cities for four centuries. A standard cross-city regression says a century of guild presence goes with a city twice as large. Against cities that were the same size beforehand the premium is about 14 percent, and the gap between the two is selection. Rights that rulers granted, such as market rights and town charters, went to towns no bigger than their neighbors. Two questions predict how selected the places holding an institution will be. Does it pay only once a place is big, and how many places can have it? The 14 percent is an association. This is work with Eric Wilhelm.

Read the whole thread

1/ Did craft guilds help medieval cities grow or hold them back? Economists have argued about it since Adam Smith. Guild cities were bigger. They were also twice the size of other cities four centuries before their guilds first appear in the record.

2/ Ghent grew from 5,000 in 1000 to 64,000 in 1300, third north of the Alps after Paris and London. Its guilds first appear in the record in the 1200s. By 1600 it had lost half its people. Read naively, guilds arrived and the city declined. The record caught them near the peak.

3/ We extended Sheilagh Ogilvie’s guild database to 1,754 European cities, 1000-1800. Craft guilds are documented in 724 of them, each dated by its first record.

4/ A standard cross-city regression on these data says a century of guild presence makes a city about twice as large. Compare guild cities with cities that were the same size beforehand and the premium falls to about 14 percent. The gap is selection.

5/ Ghent’s decline is consistent with reversion to the mean. A city that clears the size bar near its peak owes part of its size to good shocks that fade. Only against cities of comparable size does the 14 percent guild premium persist.

6/ Is this special to guilds? We ran the same comparison on eight other institutions. Rights that rulers granted went to ordinary places. German towns that received a market right or a town charter were no bigger than their neighbors before the grant.

7/ What explains this difference? Most towns are small and a few are huge. If an institution pays only above some size, only the tail gets it, and the tail averages far above the typical town. If only a few places can have it, they go to the very top.

8/ So two questions predict how selected institutions will be. Does it pay to create it only once a place is big? How many places can have it? Capitals and archbishoprics, which few cities held, sit near 3x the average. Rights most towns got sit at 1x.

9/ The sharpest contrast is inside Germany. Guilds and town charters were about equally common there, about two towns in three. Guild towns were ahead of the average town before the record and charter towns were not. Exclusivity does not separate them. The size bar does.

10/ The record also dates them differently. A market grant is written down the day it exists, so its first mention dates a birth. A guild gets written down once it is worth confirming, taxing, or fighting over, so its first mention dates an ascent.

11/ The same problem runs through policy today. Occupational licensing is the modern guild. Before copying an institution from the places that built it themselves, ask whether your place clears the bar those places had already cleared.

12/ Menger called the two origins organic and pragmatic. Hayek called them grown and made orders. Nobody designs a grown order, so nobody documents its design. Hayek’s knowledge problem is an archivist’s problem too. Which kind works better is a separate question.

13/ What we cannot say. The data do not tell us whether guilds caused city growth. The 14 percent is an association. Market rights and charters are observed only for German and Austrian towns, so that comparison is narrow and its intervals are wide.

14/ Populations are Buringh’s. Paper with Eric Wilhelm, with an interactive companion. /end

One city, concretely. Notable people born in Antwerp. 1 in the 1300s, 17 in the 1400s, 309 in the 1500s. In our data its guilds enter the record in the 1400s, as the rise becomes visible.

The second measure. Among cities of the same size, guild cities had about twice the notable births in the four centuries before their guilds were recorded, so the lead is prominence as well as size.

Capitals show the second question at work. Future capitals were ordinary four centuries before the court came, 3.6x other cities a century before, and 6x a century after. A scarce prize went to whichever city had just surged.

Method. A panel of 1,754 European cities from 1000 to 1800, with populations from Buringh and guild records that extend Sheilagh Ogilvie’s databases. Each institution’s cities are compared century by century with cities in the same pool that never acquired it. The 14 percent comes from a stacked event study that matches guild cities to never-documented cities of the same size and guild strength. Market rights and town charters come from the Deutsches Städtebuch as coded by Cantoni, Mohr, and Weigand. Notable births come from Laouenan and coauthors (2022).

The paper, Selection in the Institutional Record, has an interactive companion where you can move the two levers yourself.

The model learns queen from 500-year-old titles

The classic word analogy asks the model for king minus man plus woman. Run on embeddings trained only on early-modern Latin book titles, the top answer is regina. The queen word also appears in the top ten for English and French and Dutch. The embedding spaces are small and noisy and trained on short titles so personal names bleed in. The structure is there at all in text this old and this sparse.

Read the whole thread

1/ The most famous party trick in AI: take word vectors and compute “king − man + woman.” You (often) get queen — evidence the model learned gender as a direction in space. Could you pull that off using nothing but the titles of books printed in Europe, 1450–1650?

2/ I trained word embeddings on ~800,000 early-modern book titles across seven languages and ran the analogy in each. Latin is textbook-perfect: rex − uir (man) + femina (woman) → regina (queen) — the single closest word. From book titles. From 500 years ago.

3/ And Latin isn’t alone. Queen lands in the top handful in three languages:

• Latin: rex − uir + femina → regina (#1)
• English: king − man + woman → queene (#3)
• French: roi − homme + femme → reine (#4)

(Dutch’s koninginne sneaks in at #8, just outside.)

4/ Where it “fails” is just as revealing. Spanish doesn’t return reina — it returns Felipe, Filipe, Phelipe (King Philip) and “III, IV.”

5/ And the woman direction keeps surfacing actual women — the neighbors fill up with names: Elisabetha, Christina, Louise, Sybilla — plus French damoiselle (damsel) and veuve (widow).

6/ Caveats for the pedants (I love you): these are small, noisy vector spaces built from short titles — cosine similarities ~0.3–0.5, not modern-LLM clean. Italian even fell over because “re” (king, two letters) got filtered out of the vocabulary.

7/ Still — a gender axis, recoverable from the semantic geometry of 500 year old book titles, cleanly enough that Latin hands you regina. The market for ideas had structure. The vectors can see it. /end

The paper behind this essay has an interactive companion.

The word for witchcraft pointed at a different fear in each language

Everyone in early modern Europe feared witches and they did not fear the same thing. The word for witchcraft points somewhere different in each language. Latin points at a book, the Malleus Maleficarum and the treatise vocabulary around it. German points at the weather, at hail and tempests. French points at devils and possession. English points at sin and at women. Italian and Spanish and Dutch have no vernacular witch word that clears the bar, only the devil. In Italy and Spain the witch stayed in Latin and in the Inquisition’s books. The Dutch absence needs a different explanation.

Read the whole thread

1/ Everyone in early modern Europe feared witches — but they didn’t fear the same thing. Feed 200 years of book titles (1450–1650) into a word-embedding model, and each language’s word for “witchcraft” points somewhere different.

2/ Latin (maleficarum): the nearest word is malleus — the Malleus Maleficarum, the 1487 witch-hunting manual. The rest is treatise boilerplate (“divided into parts, tomes”). Witchcraft = a learned book. The scholar’s witch.

3/ German (zauberey): hagel (hail), wetter (weather), ungewitter (tempests). Weather magic. The peasant’s witch — she conjures the hailstorm that ruins your harvest.

4/ French (sorciers): diables, démons, anges, possession. Pure diabolism. The exorcist’s witch — spiritual warfare, the world of the great possession cases.

5/ English (witchcraft): sin, idolatry, pride, disobedience — and women. The preacher’s witch: a species of sin, gendered female.

6/ Same word, four fears — a treatise, a storm, a demon, a sin. Each matches what historians know about those regions. The model read it straight off the titles.

7/ But the best part is where the witch isn’t. Italian, Spanish and Dutch titles have no vernacular word for witchcraft that clears the bar — only the devil. In the Catholic south the witch never went vernacular; she stayed in Latin, in the Inquisition’s books.

8/ And even the leftover devil differs: in Italy a tempter (hell, temptation), in the Dutch Republic a devotional foe (scripture, flesh, world), in Spain a stage villain (comedia, chivalric romance). In Spain, the devil went to the theater.

9/ Coda for the curious: the German “weather-witch” isn’t just linguistic — economists have shown severe weather actually predicted witch-trial spikes (Oster 2004).

10/ And on why the trials finally ended, Mark Koyama and I have written on France — as the state built fiscal and legal capacity, its courts reined in witch prosecutions. /end

The paper behind this essay has an interactive companion.

Every city printed something different

Take the average embedding of everything a city printed in Latin and subtract the European average. The words nearest that distinctive direction are the city’s print personality. Lyon was law. Louvain was music. Venice was marketing, in the sense that its distinctive words are the publisher’s blurb. Wittenberg was the Reformation and Ingolstadt its Catholic mirror image. This is Latin only so it captures a city’s international signature. Some cities returned only generic university vocabulary and are left off the map.

Read the whole thread

1/ Every city that printed books in early-modern Europe had a “personality”. A few words that set its output apart from everyone else’s. We measured it using word and title embeddings for hundreds of cities, straight from 300,000 Latin book titles between 1450 and 1650. Some are really cool…

2/ Start with Louvain. Average all its Latin titles in “meaning-space,” subtract the European average, and the leftover words read like a title page — “motets,” “of songs,” “the soprano part,” “of voices,” “for five.”

3/ Why? Louvain was home to the Phalèse press. Pierre Phalèse, one of the 16th century’s great music publishers, printing motets and chansons for four, five, and six voices. The model rebuilt his catalog without knowing music exists.

4/ Now Venice. Its signature is stranger. Not a subject at all. “Newly.” “Just recently.” “With additions.” “Most diligently.” “A new edition.” “With summaries.” Venice’s fingerprint isn’t what it printed. It’s the sales pitch.

5/ History again. Venice was Europe’s most competitive print market and it built the rules to match. It granted the first printing privilege in 1469, an early copyright, and passed Europe’s first patent law in 1474. Its title pages read like ads because they were.

6/ The rest of the map tells the story for a few other cities. Lyon prints law. Wittenberg prints the Reformation. Each city’s distinctive words are a snapshot of what it sold to Europe.

7/ The Method in one breath. Each title is the average of its word embedding vectors. Then we take the average of the city’s titles minus the European average. Then the remaining “unique” city signature is the nearest words. /end

The paper behind this essay has an interactive companion.

Women in the printing trade

Women appear in the corpus as printers rather than authors, and overwhelmingly as widows continuing a husband’s press. They account for just under two percent of editions, concentrated in the Low Countries and central Germany and the Baltic and nearly absent from Italy and Iberia. Presses run by women look more institutional in what they print. But a before-and-after comparison within each press shows widows printed almost exactly what their husbands had. The gap is selection over which presses passed to widows rather than a change in behavior, and the passing itself happened where law and custom let a widow keep the business. Female authorship is not observable in this extract of the catalogue, so its absence here is a data limitation rather than a finding.

Read the whole thread

1/ In some Renaissance cities, women printed as many as 1 in 5 of all books. In Italy, almost none. Here’s the map, and it traces one of the deepest divides in European history. 🧵

2/ These women weren’t authors, almost no woman published under her own name during this period. They were the printers, the businesses. And they got there one way, a master printer died, and his widow took over the shop.

3/ The imprints say it outright, “widow of Christophe Plantin,” “widow of Jan Cnobbaert.” Some ran the great houses: Charlotte Guillard in Paris, Katharina Gerlach’s music press in Nuremberg, the widow of Blas de Robles printing the crown’s edicts in Madrid.

4/ And an unexpected fact, female printers printed exactly what their husbands did — same law, dissertations, edicts, music. Across the presses, before vs. after the handover, the output barely moves. The widow kept the business running.

5/ Often she was the last of the line. In about half of these presses a son or heir of the same name resumed printing after her. In the rest, the press ended with her. Either way she carried the family firm across the founder’s death.

6/ So why the North-South gap? Line the countries up and it’s stark. A woman printed roughly 1 book in 22 in Belgium — and about 1 in 1,700 in Italy. In the North Sea region women married late, worked for wages, and could run a business as a widow. Southern Europe much less so. This map of printers is really a map of women’s economic rights.

7/ Economists have a name and an origin for this. Voigtländer and Voth trace it to the Black Death — women went to work, married later, gained standing; De Moor and Van Zanden call it “girl power.”

8/ It comes down to whether the right to run a business existed at all — the deep gender divergence showing up in who printed Europe’s books. Women shaped the early book trade less as writers than as the people who kept the presses alive. /end

Method. USTC, about 680k editions across 7 languages, 1450 to 1650. I flagged female run imprints, mapped the city shares, and compared each press’s output before and after the widow took over.

The moral standing of commerce, one language at a time

Deirdre McCloskey argues that modern growth required a revaluation of commerce from sin to dignity. The embeddings let you read that gradient across languages in one figure. The neighbors of the Latin word for trade are usury and simony and homicide. That is the vocabulary of canon law. The neighbors of the Dutch word are free and company and charter. English and French sit near the honorable end and Spanish near the suspect end. Every token in the figure is a genuine nearest neighbor with its real cosine similarity. The one word labels are my reading and are marked as such.

Read the whole thread

1/ In 500-year-old book titles, I asked one question. What words sit closest to each language’s word for trade and the merchant? In Latin, trade keeps company with usury, simony, homicide. In Dutch, it’s free and chartered. In between lies the whole moral history of capitalism. 🧵

2/ How? A word embedding learns a word from the company it keeps. Feed it about 800,000 book titles printed across Europe, 1450–1650 (the Universal Short Title Catalogue), and words that share similar titles land near each other. So I can ask what “trade” sat beside — language by language.

3/ LATIN — the tongue of the schoolmen and canon law. “mercatura” (trade) sits with usura (usury), simonia (simony), even homicidio (homicide), and restitutio (restitution). Commerce is a sin to be confessed and paid back.

4/ SPANISH — the crown’s proclamations. “comercio” sits with rebelde (rebels), prohibición (bans), the almirantazgo (admiralty), and olanda (Holland — the enemy). Trade is contraband — licensed, embargoed, shipped past the foe.

5/ ITALIAN — “mercanti” sit among foro (the courts), statuti (statutes), dogana (customs), sensali (brokers), banchieri (bankers), and soria (Syria — the Levant routes). Commerce as regulated machinery.

6/ FRENCH — “marchans” sit with échevins (aldermen), prévôt (provost), citoyens (citizens), consuls, and libertez (liberties). Now the merchant is a civic figure who holds franchises and rights.

7/ ENGLISH — “merchant” sits with alderman, citizen, worshipful, and the Adventurers — the great chartered companies. A guild brother of a worshipful company. Commerce has become an honor.

8/ DUTCH — merchant Amsterdam. “negotie” sits with open and vry (free), compagnie (the VOC/WIC), octrooi (charter), Brazil and the West Indies. Trade is free, chartered, and global — and a nearby word, profijtelijck, means profitable.

9/ Read top to bottom, commerce turns from a sin to be confessed into a freedom to be defended. That is the “bourgeois revaluation” Deirdre McCloskey argues made the modern world — here it is, written into 500 years of book titles.

Method. Word vectors are PMI plus truncated SVD on titles. Neighbors are cosine nearest. Curated only to drop OCR and place name noise. Every token shown is a genuine top 20 neighbor. One honest note on the comparison. The contrast is between registers, clerical Latin against merchant Dutch, rather than a timeline. That is the point.

The paper behind this essay has an interactive companion.

What the Thirty Years’ War did to German print

Total print output inside the Holy Roman Empire held up through the war and even grew. Its composition collapsed. Useful knowledge, meaning theology and law and medicine and science and history, fell by roughly forty percent while news and occasional print nearly tripled. The press shifted from books to news. Useful knowledge output ran parallel inside and outside the Empire until the 1610s and then forked, which points at the war rather than a Europe-wide change in taste.

Read the whole thread

1/ You’d expect the Thirty Years’ War (1618–1648) to have wiped out German printing. It’s stranger than that. The cities that were sacked or besieged saw their presses collapse — but neutral cities boomed. The trade didn’t die. It moved. 🧵

2/ Magdeburg, sacked in 1631, fell from about 430 German editions a decade to 40. Neutral Hamburg rose about 26-fold, to nearly 2,000. And it isn’t just those two — across the sacked cities (Augsburg, Heidelberg, Frankfurt/Oder) presses collapsed; across the neutral ones (Köln, Gdańsk, Zürich) they soared.

3/ But here’s the stranger thing. Zoom out to the whole war zone and total print didn’t fall at all — it grew. What collapsed was what they printed.

4/ Useful knowledge — theology, law, medicine, science, history — fell by about 40%. Meanwhile news, periodicals, and propaganda nearly tripled. Useful knowledge went from a third of war-zone print to a sixth. The presses ran hot. They just stopped making books and started making news.

5/ And this was the war, not a Europe-wide fashion. Over the same decades, useful-knowledge output grew about 46% in the rest of Europe while the German lands’ fell about 40%. Learned print drained west, to the Dutch Republic and France.

6/ Why? Real scholarship needed the universities, the Frankfurt book fair, paper, and trade routes — all wrecked by the war. News needed only a press and a rumor. So the presses stayed busy, and the knowledge did not.

7/ The honest caveats. These are surviving, catalogued editions; the classifications are coarse; “war zone” is the Holy Roman Empire, a proxy. Descriptive, not causal. But the pattern — output held, useful knowledge hollowed out — is striking, and it says something real about how war reshapes the market for ideas. /end

Method. Editions by USTC classification, grouped into useful knowledge against news and occasional print, with the war zone proxied by the Holy Roman Empire. Edition counts of surviving books.

Districts near the 1857 reprisals vaccinated less for decades

This essay draws on my paper with Malik Altaf Hussain about vaccine hesitancy in colonial India. Before refrigeration the smallpox vaccine was a live virus with no shelf life, so the British moved it across India through the arms of children, vaccinating one and harvesting the lymph a week later for the next. The campaign saved lives and it also arrived by force, from a state that had recently been very violent. Districts closest to the reprisals that followed the 1857 revolt vaccinated less for decades afterward, and the effect fades as the generation that lived through 1857 passes. State violence carries a lasting public-health cost.

Read the whole thread

1/ Before refrigeration, the smallpox vaccine was a live virus with no shelf life. So how did the British move it across 19th-century India? They chained it through the arms of children. Vaccinate one, wait a week, harvest the lymph from the blister, use it on the next. 🧵

2/ To the medical officers running the program this was just efficient epidemiology, the fastest way to keep a live vaccine alive. But it meant taking villagers’ children from place to place. And the colonial government’s own political leadership saw the danger the doctors didn’t.

3/ In the Punjab government’s review of the 1868-1880 report, the Lieutenant-Governor overrules his own Superintendent-General. The doctor saw “no great hardship.” The political office called the practice “certainly unlawful” and warned it would “render vaccination unpopular.”

4/ A tension that outlives the 19th century. You can get the science of saving lives exactly right and still get the politics of it wrong. The technical instinct optimizes the medicine and leaves the harder question, whether people trust the hand delivering it, for someone else.

5/ They were also displacing a local practice. Indians had variolated against smallpox for generations (deliberate mild exposure, tied to the goddess Sitala). The state pushed its own vaccine, made it compulsory in cities (Bombay, 1877) and in 1880 banned the Indian method.

6/ And where the state had recently been most violent, the friction was worst. In Lucknow, epicenter of the 1857 reprisals, the 1873 vaccination report records a vaccinator dismissed for trying to vaccinate a bazaar “against their will.” Sixteen years on, force still met refusal.

7/ In our research, districts closest to those 1857 reprisals vaccinated less for decades. The vaccine worked, and smallpox was merciless. But a life-saving technology can still fail when it arrives by force, from a state you have every reason to distrust. /end

The paper is Hussain and Johnson, The Great Revolt and its Legacy, Understanding Vaccine Hesitancy in Colonial India. The archival quotes are verified against the original scans. The route map is adapted from Bennett (2007).

The paper behind this essay has an interactive companion.