Run of the mill

There is broad agreement that human capital — the skills and attribute which influence individuals’ productive capacity — matters for long-term development. But curiously, there is mixed evidence on the role of human capital as a driver of industrialisation.

Part of the problem here may relate to how human capital is measured. At the macroeconomic level, human capital is typically measured with respect to education levels across the population: for example, years of schooling or literacy rates. But this may be less relevant than the ‘upper tail’ of knowledge in society: the select few with highly specialised skills. Mokyr, Sarid and van der Beek (2022) test this idea of upper-tail human capital in eighteenth-century Britain. Specifically, they look at the role of watermills, and the millwrights who were responsible for them.

The central question the authors consider is how the geographic distribution of millwrights — those possessing an important and specialised skillset — influenced the location of industry. To answer this, they draw on district-level data from England between 1710 and 1750, measuring employment by occupation, production activity and geographic factors. These data are coupled with medieval records of the location of mills, contained in the eleventh-century Domesday Book.

The Mill at Dorking by Henry Hewitt, 1869. Source: Wikimedia Commons.

Centuries apart

The first stage of analysis is a straightforward ordinary least squares (OLS) regression, examining the effect of medieval-era watermills on the number of millwright apprentices in the early eighteenth century. The authors find a significant and positive correlation between the two: on a per capita basis, an additional watermill in the eleventh century is associated with an increased number of wright apprentices in the eighteenth century by around 0.15. A suite of different controls are included — such as latitude/longitude, soil quality, and climate — which do not materially affect this headline finding. Furthermore, as shown below, two-stage least squares (2SLS) estimates — using geographic factors as an instrument — are also fairly consistent with the baseline OLS results.

The effect of medieval mills on 1700’s millwrights

Dependent variable: Wright apprentices per capita, 1710–1750OLS
Domesday Book watermills (per capita)0.15***
River suitability2.28
Wheat suitability1.18*
Main controlsNNYYYY
Additional geographic controlsNNNNYY
Adjusted R20.570.570.590.590.620.59
Per capita estimates controlling for distict population. 2SLS estimates report only second-stage results. Main controls include longitude, latitude mean ruggedness, mean elevation, agricultural suitability, suitability to grow pasture and wheat, mean precipitation and temperature. Additional geographical controls account for the distict’s area as well as the distict’s distance from London, main eighteenth-century harbours, historical Roman roads and navigable rivers. Robust standard errors in parentheses.
Source: Table 3, Mokyr, Sarid and van der Beek 2022.

Additional analysis looks at the effect of the Black Death in the fourteenth century, which prompted several mills to close: due both to a direct effect from the massive decline in population, and an indirect effect from population-induced changes in real wages motivating changes in farming practices. (In this period, mills were used to process grain. After the Black Death, crop farming lost ground to pastoral farming — in particular, sheep.) Those areas which, on a per capita basis, recorded the greatest fall in mills due to plague-induced population declines in turn had fewer millwrights in the eighteenth century.

These results imply a degree of path dependence, which is not especially surprising. For centuries, mills were an important technology, and a technology that required certain skills. That there should be a correlation between the location of mills and the location of relevant skilled workers — even over long time periods — is of itself unremarkable. But what happens when technical advances change the underlying structure of the economy?

The past is prologue

Mokyr, Sarid and van der Beek use their initial analysis of persistence as springboard to consider how mills and millwrights influenced productive activity in Britain’s industrialising economy.

The authors estimate the relationship between the number of medieval mills and the number of apprentices in a range of different professions — only some of which were at the leading edge of mechanised activity in the eighteenth century. Equivalent to the table above, the number of medieval mills is instrumented by river and wheat suitability. As indicated in the following table, there is a significant effect on the number of apprentices in the textiles and iron-making industries (columns 1-4) — among the first to industrialise — but not for other occupations.

The effect of medieval mills on apprenties in textile manufacturing, iron-making and other fields

Dependent variable: Apprentices in given occupation (columns 1–10) per capita, 1710–1750

Domesday Book watermills (per capita)0.12***
Wheat suitability0.25
River suitability10.75
Adjusted R20.310.170.550.640.410.030.600.510.510.38
2SLS per capita estimates controlling for distict population; second-stage results only. The full range of controls as described in the preceding table (Table 3, Mokur, Sarid and van der Beek 2022) are applied to all regressions shown here. Robust standard errors in parentheses.
Source: Table 7, Mokyr, Sarid and van der Beek 2022.

The results above are indicative of a clustering effect: that those areas with a historic concentration of mills (and millwrights) were also those where early industrial activity was typically located — even when controlling for the geographic factors that likely mattered. But this analysis does not of itself explain whether it was proximity to mills (an infrastucture and energy story) or proximity to millwrights (a human capital story) that mattered more.

To examine this aspect, the authors perform a ‘horse race’ between the effect of both mills and millwrights on textile manufacturing and iron-making (measured by the number of apprentices in those industries). Not only is the magnitude of the effect of millwrights greater than the effect of mills when the two are tested separately, but much of the direct effect of mills on industrial apprentices comes from (is ‘mediated’ by) the number of millwrights.

What matters most: millwrights or mills?

Dependent variable: Apprentice drapers/smiths per capita



Wrights (per capita)0.38***
Domesday Book watermills (per capita)0.06**
Mediation percentage68%49%
Adjusted R20.450.390.450.650.600.68
Selected results (columns 1–3, 7–9) shown here. The full range of controls as described in the preceding table (Table 3, Mokur, Sarid and van der Beek 2022) are applied to all regressions shown here. Robust standard errors in parentheses.
Source: Table 8, Mokyr, Sarid and van der Beek 2022.

It is worth emphasising that these results only identify correlation; it would be too strong to conclude that millwrights ’caused’ industrialisation. The argument here is that the relationship between millwrights and industrial activity in eighteenth-century England highlights a role for human capital. The local availability of specialised workers with mechanical skills had long origins: centuries of experience with watermills had resulted in a skilled workforce that was useful for early industrialisation.

Does a wheel work differently in Britain?

Mokyr, Sarid and van der Beek conclude by discussing how ‘human capital’ should be conceptualised — not in terms of years of schooling and formal institutional settings, but with a focus on the skills acquired by individuals. This is true enough. But I’m not convinced that their analysis gets to the heart of skills acquisition.

This paper focuses on a group of workers with a key set of skills. It is a labour story rather than a human capital story, per se. We do not observe the skills attained by millwrights or their industrial contemporaries. While it is certainly plausible that the presence of a skilled workforce mattered for industrialisation, we cannot say precisely which skills mattered. Nor is there direct evidence on the various channels for skills transmission.

Getting into the mechanics. Source: William Emerson / Wikimedia Commons.

Watermills — and millwrights — were not unique to Britain. But it was Britain that was first to enter the industrial age. The authors do not directly posit that upper-tail human capital was key to this. But if it were, this would imply some skill difference between Britain and (for example) the rest of Europe. Did the capabilities of British millwrights and related craftsfolk substantially differ from their continental peers? And if so, why?

There is a brief suggestion from the authors that the British approach to technical advances was more practical than elsewhere in Europe, with a greater focus on nuts and bolts rather than scientific theory. Perhaps this is correct, but it is hard to validate. (The appendix material notes one view among the scientific elite of the day that British millwrights were little more than ‘glorified carpenters’. But one could imagine hearing similar remarks within France or the Netherlands with respect to their own mills.)

To the extent that there were differences between Britain and the rest of Europe, these may have less to do with the technical content of skills than their application. Were British millwrights simply better at employing new technologies than their continental peers? Did their skills adjust more quickly in response to technical advances? An ability (or indeed willingness) to adapt to change perhaps is the real driver — and this need not be concentrated among the upper tail.

That said, Mokyr, Sarid and van der Beek make a strong case for focusing on the upper tail of knowledge rather than the broad masses. And pre-industrial millwrights are a relevant demonstration of the idea, which the authors richly mine for useful insights. Still, their analysis inescapably invites further questions. One thing is knowing who had skills; it is something else to identify what those skills were and how they were used. Upper-tail human capital may be the right concept, but its measurement still requires refinement.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.