Education — whether through formal instruction in schools, or through on-the-job training — is essential in the development of skills. And skills matter for workers’ productive capacity: that is, how we apply ourselves in our working lives. While there is a considerable body of research on the role of education in raising labour quality and output, there is much that remains unknown — or at least empirically unverified. One of the big question marks relates to the effects of different types of skills and education.
In a recently published paper, Diebolt et al. (2021) offer a fresh perspective on how industrialisation influenced the demand for skills. They consider specific types of education in nineteenth century France, and show a relationship between the rise of the steam engine and the accumulation of practical skills with technical applications.
The authors contrast ‘basic’ and ‘intermediate’ skills. Basic skills can be thought of as literacy and numeracy obtained through primary schooling. Intermediate skills build (naturally enough) on the basic skillset, relating to general competencies in fields such as applied natural sciences, foreign languages, law and commerce. These intermediate skills must nevertheless be distinguished from advanced, specialised skills conveyed through (for example) vocational education and training.
The authors focus on two types of educational institutions which were launched in France in the early- to mid-1800’s: the Ecoles Primaires Supérieures for 12–16 year olds, and the Cours d’adultes et d’apprentis, which was a type of adult education to train workers. Though their students differed, both types of institution had a similar purpose in terms of developing intermediate skills, with a focus on the working and middle classes (in contrast to the elite with access to secondary education).
The methodological engine room
The broad research question considered by Diebolt et al. is, did industrialisation drive the acquisition of intermediate skills in nineteenth century France? As a proxy for industrialisation, they use the number of steam engines by county between 1839 and 1900. Steam engines had been regulated in France since the 1820’s: that is, all installations of steam engines had to be approved. These approvals were reported in statistical records, which the authors draw on for their analysis.
As the authors report, the number of steam engines increased significant during the century: from (in round figures) 2000 in 1839 to 75,000 in 1900. Moreover, there were significant differences across counties — both in engine numbers and their growth over time. It is these differences which the authors exploit to assess effects of industrialisation on education.
As a measure of skills, the authors consider enrolment numbers at the intermediate educational institutions described above. In addition, to test the effect on basic skill acquisition, they include various measures related to literacy and primary schooling. As such, they can compare the effect of industrialisation on intermediate skills relative to the effect on basic skills.
The authors also apply various controls to exclude the effect of likely confounding factors. Included in these are measures for infrastructure, population density, and geographical attributes (land quality, climate, sea access). Moreover, the authors control for pre-industrial skill levels, using a measure of literacy using marriage certificate signatures. (One can well argue that the ability to sign a document is a poor measure of literacy — see, for example, Nilsson (1999). Nevertheless, as an approximation of the relative differences across counties, it is probably an acceptable benchmark.)
Diebolt et al. present results from several different model specifications. The first series of results, using OLS and fixed effects models, shows a strong, significant correlation between number of steam engines and enrolments in intermediate education.
Moving from correlation to causation, the authors employ a two-stage least squares (2SLS) analysis to address possible endogeneity (for example, that the rise in education might accelerate industrialisation). Their chosen instrument for this task is distance between each county (measured from its capital) and the northern town Fresnes-sur-Escaut, where the first French steam engine was installed. (As distance to Fresnes increases, the number of steam engines per county on average decreases: a relationship that holds throughout the time period considered.)
Like any application of instrumental variables, it is reasonable to question whether the chosen instrument is valid. Diebolt et al. motivate their choice with reference to a study by Franck and Galor (2021), which also explores the relationship between steam engines and education in France — though in the case of that study, the focus was education in terms of literacy and schooling. Franck and Galor offer a more detailed discussion than Diebolt et al. of the history of the steam engine in France, better clarifying how the technology spread from Fresnes to neighbours in the north of France. Robustness checks show that the number of steam engines in a county was not correlated with that county’s distance to Paris or other major economic centres in France.
The results of the 2SLS estimates confirm the general finding that the number of steam engines is positively associated with enrolments in intermediate education. As the table below shows, the estimates with respect to intermediate adult education — Cours d’adultes et d’apprentis — lose significance once controls are applied. However, the authors recover the effect when they focus on enrolments in the second half of the century, when the gains from education with respect to industrialisation are greatest.
|intermediate (youth) education||intermediate (adult) education|
|No controls||All controls||No controls||All controls||All controls (post-1863)|
|Number of steam engines||0.505***||0.593**||0.436***||0.151||0.402**|
|Distance to Fresnes||−1.149***||−0.857**||−1.478***||−0.932***||−0,874***|
Thereafter, the authors run cross-sectional regressions for three years — 1839, 1861 and 1886 — to compare how the effects of industrialisation on education change over time. The effect on enrolments in intermediate education only becomes significant from 1861, and with a greater magnitude of effect evident in 1886. By contrast, the reverse is true for the effect on measures of basic education: their significance falls across the three observed years. This is consistent with the story that industrialisation (that is, the rising number of steam engines over time) motivated greater take-up of specifically intermediate education, rather than education across the board.
Steaming ahead, but why?
One thing is to establish a relationship; another is to explain the causal channel(s). Diebolt et al. make clear that their analysis does not attempt the latter. Nevertheless, in discussing their results, they point to two plausible (and by no means mutually exclusive) explanations of the observed effect.
- Industrialisation resulted in rising household wealth. This wealth both generally increased the demand for education, but also changed household expectations and aspirations — families began to a see a future for themselves (or rather, their children) that motivated investment in intermediate skills.
- The technical changes underpinning industrialisation increased the private returns to technical skills necessary to, for example, build, maintain and operate increasingly complex machinery. Investment in intermediate skills became households’ entry ticket to sharing in the gains from industrialisation.
As the authors state, a more detailed analysis would be required to test these causal channels, and disentangle the relative strength of each channel.
What the analysis doesn’t shed light on is the reverse effect of education on industrialisation. That is, to what extent did different types of education support the technical changes that were essential to the industrialisation process? It is reasonable to suppose that growth in more technically oriented education also expands the base of skills and knowledge from which innovations in products, materials and techniques emerge. It is, however, no easy task to establish this empirically. It would in any event require a much larger range of data (particularly with respect to different aspects of industrialisation) than Diebolt et al. have to hand.
In conclusion, Diebolt et al. offer a welcome contribution in unpacking the relationship between economic development and education. The novelty of their analysis is the focus on a particular type of education: that is, the teaching of intermediate skills. Much economic analysis in the field of human capital considers education in aggregate: that a year of primary school is (very crudely speaking) interchangable with a year of technical training. This oversimplification is plainly untrue, and misses the point that education involves the development of many different productive skills. There is considerable untapped potential when it comes to disaggregating these differences: moving away from aggregate constructs of years of schooling to understanding what skills individuals acquire, and how those skills relate to long-term economic growth.