Nine centuries of cultural change and the productivity of ideas

Economic historians have long placed culture at the centre of the rise of the West, but have lacked a measure of culture spanning the emergence of modern growth. Using 23,064 historical books, this column constructs an ‘innovation wedge’ that separates the values texts endorse from the social arrangements they merely depict. The wedge falls by 51% between 1000 and 1920. In a calibrated growth model, that decline accounts for about two-thirds of the first sustained acceleration in productivity growth between 1500 and 1700.

Productivity growth has slowed across advanced economies, and one prominent diagnosis is that research productivity itself is falling. More researchers are needed to sustain a given rate of technological progress (Webb et al. 2017, Bloom et al. 2020). Economists can readily measure inputs such as R&D spending and scientific talent. The efficiency with which those inputs become ideas is harder to see.

Culture has long been part of the answer to what made modern growth possible. Weber (1904-05) linked ascetic Protestant values to the spirit of modern capitalism. Mokyr (2016) links modern growth to changing attitudes towards useful knowledge between 1500 and 1700, while McCloskey (2016) places a new respect for commerce and innovation at the heart of the Great Enrichment. Henrich (2020) takes the story further back, tracing Western individualism to the Church’s long reshaping of kinship and social organisation. Economists have formalised how cultural traits pass between generations (Bisin and Verdier 2001), but have lacked a measure that follows cultural change across the centuries in which modern growth emerged.

Quantitative work on texts has a longer history in the digital humanities (Moretti 2005, Piper 2018, Underwood 2019). Recent work in economics has begun to measure aspects of culture using very different kinds of evidence. Michalopoulos and Xue (2021) use coded oral traditions to recover cross-sectional differences in culture across roughly 1,000 pre-industrial societies. Galor et al. (2023) study 2,564 motifs from 958 oral traditions to trace the roots of cultural diversity. Voth and Yanagizawa-Drott (2024) use more than 14 million US high-school yearbook portraits to track style individualism, persistence and novelty from 1950 to 2010. None of these measures spans the emergence of modern growth. In a recent paper, I use historical books to build a measure across the missing centuries (Stefanski 2026).

Reading the canon, not counting its words

I assemble 23,064 books in English, including translations, from Project Gutenberg, a public-domain library. I date each book by the composition of the underlying work rather than the edition, and end the series in 1920. In the paper, I use a language model to read passages from each book in the context of the whole work. The model judges whether each text endorses, rejects or merely depicts positions along six dimensions adapted from cross-cultural psychology (Hofstede et al. 2010): deference to hierarchy, fear of the untried, individualism, patience, competitive striving and indulgence.

The distinction between endorsement and depiction is the heart of the method. A conventional dictionary-based measure of power distance gives Machiavelli’s The Prince a hierarchy score of 66.7 and Venture Smith’s memoir of enslavement a score of 100. It sees the language of command and ownership in Smith and mistakes the hierarchy he describes for one he approves. The contextual reading reverses the ranking, 100 for The Prince and 0 for Smith. Machiavelli treats princely authority as an instrument of political action; Smith depicts domination while condemning it and asserting self-ownership.

Fourteen scholars in literature, classics and modern languages at St Andrews compared books they knew without seeing the model scores. Their judgments agreed with the model in 93% of comparisons where they knew both books and chose a direction. If “about equal” and “cannot judge” are counted against the model, the agreement rate falls to 65.5%. Five of the six measures correlate in the predicted direction with pooled modern cross-country benchmarks in Europe and North America; power distance is the exception.

Culture is inherited rather than reinvented each year. I therefore accumulate book scores into slow-moving stocks, using a 0.5% annual depreciation rate, which gives a half-life of about 138 years. Figure 1 shows how the six inherited stocks change over time.

Figure 1 Six cultural dimensions measured in the historical book corpus

Note: Inherited stocks on a 0-100 scale. The shaded pre-1000 segment has sparse textual coverage and is shown only for context.

Nine centuries of falling resistance

Between 1000 and 1920, hierarchy and fear of the untried decline, while individualism and patience rise. Competitive striving also falls and indulgence rises. The rise in patience complements Stefanski and Trew (2024), where greater effective patience supports accumulation; here it works through the productivity of research.

Much of the movement begins before industrialisation. It is not an artefact of later books going silent on the six dimensions. Passages take a clear position about as often in the nineteenth century as in the eleventh. What changes is the position they take.

I combine the four dimensions most directly tied to idea production into an ‘innovation wedge’, with hierarchy and fear of the untried in the numerator and individualism and patience in the denominator. A broader companion, ‘total friction’, adds competitive striving and indulgence to cover all six dimensions. A high innovation wedge means more cultural resistance to dissent, experimentation and patient accumulation. Between 1000 and 1920, it falls by 51%.

The decline survives several alternative ways of combining the six dimensions. Applying the method unchanged to an independently assembled archive of 4,996 books from four other collections produces the same broad decline. The results are also similar when validation samples are rescored under different instructions and with DeepSeek, a separately trained language model. Figure 2 compares their paths.

Figure 2 The innovation wedge falls by 51% between 1000 and 1920

Note: Total Friction, the broader six-dimensional measure, falls by 28%.

From measured culture to measured growth

I embed the wedge in a semi-endogenous growth model in the tradition of Jones (1995). A lower wedge raises the productivity of research, so the same research effort produces more ideas. To estimate the link between culture and research productivity, I construct the same cultural stocks separately for 13 countries, assigning each book to its place of writing under modern borders. I then relate changes within each country to the number of people born roughly a generation later who went on to become notable innovators. A stricter timing check uses only the preceding 25 years of the wedge and gives a similar estimate. With only 13 countries, the estimate is informative rather than definitive.

The cultural ground starts moving well before the familiar economic take-off. Between 1500 and 1700, model-implied annual productivity growth roughly doubles, and the model attributes 64.8% of that rise to the falling wedge. This early acceleration falls squarely in the period where Mokyr locates the cultural origins of modern growth. Population scale, the engine of long-run growth in these models, becomes the larger contributor only after 1800. By 1920, holding the wedge at its year-1000 value leaves productivity at about 56% of the level reached under the measured path.

Figure 3 Annual productivity growth in the calibrated model

Note: The shaded 1500-1700 interval is the first sustained acceleration; the falling wedge accounts for 64.8% of its rise. The black line follows the measured wedge, the red line holds it at its year-1000 value, and the dashed line is the difference in percentage points.

Whose culture, and what kind of claim

The corpus captures the durable literary culture of the people and institutions that wrote, preserved and taught books. It says less about everyday beliefs or the culture of the street. For a paper about the production of ideas, this elite literary culture is the relevant object, but it should not be mistaken for Western culture as a whole.

The growth accounting assigns culture a large role, but it does not establish that cultural change caused growth. Economic change can reshape culture, and other forces may affect both culture and innovation. I would also not put too much weight on any single book score. The long-run path is more informative, and it reappears in the independent archive and under alternative scoring methods. The individual scores are still worth inspecting, and an interactive companion lets readers compare 275 classics.

The series ends in 1920 because copyright makes the later public-domain record much thinner, so it cannot diagnose today’s productivity slowdown. It does show what standard growth accounts leave out. R&D budgets and researcher counts may not explain why similar inputs produce different amounts of useful knowledge across times and places. Attitudes toward authority, uncertainty, individual agency and patience may be part of the explanation, and the books provide a way to measure them. The distinction between depiction and endorsement could also be applied to modern literary corpora, scientific writing and corporate records. Research productivity may depend on a cultural environment that allows ideas to be challenged and built upon, as well as on R&D inputs.

Source : VOXeu