The diffusion of innovation across firms is a core driver of aggregate productivity growth. Despite its importance for economic development, our understanding of technology adoption is incomplete. While a large literature has documented that new technology spreads slowly across firms, the reasons for this sluggish diffusion are not well-understood in the case of manufacturing (Rosenberg 1976, Hall and Khan 2003, Hall 2004).1 In fact, recent evidence has shown that offering manufacturing firms standard technologies can boost their productivity (Bloom et al. 2013, Giorcelli 2019). This raises the question why firms did not adopt these technologies earlier, without exogenous intervention. A second puzzling observation is that even in cases where major breakthrough technologies such as electricity and information technology diffuse across firms, the widely expected boost to aggregate productivity has failed to materialise. This observation is most famously formulated by Bob Solow, who in 1987 remarked that “[...] what everyone feels to have been a technological revolution, a drastic change in our productive lives, has been accompanied everywhere, including Japan, by a slowing-down of productivity growth, not by a step up. You can see the computer age everywhere but in the productivity statistics.”2
A natural lens to study these questions is the firm productivity distribution. A recent literature has shown that even within narrowly defined sectors, firm productivity differs substantially (Syverson 2011). Productivity differences across firms reflect both the underlying technology and the efficiency with which this technology is used. Both features play important roles in firms’ decisions to adopt new technologies: What are the potential productivity gains of a new technology, and is the operational knowledge needed to achieve these gains readily available? Exploring these features empirically has thus far proved difficult because of data limitations. Standard data sources typically do not report the technology used; and even if they did, old and new technologies often co-exist within firms, making it difficult to isolate the productivity effects of the latter. To identify the effects specific to a new technology, many recent studies use randomised control trial methods to study technology adoption in manufacturing (Bloom et al. 2013, Atkin et al. 2017, Hardy and McCasland 2016). The small sample size typically employed in these studies precludes a systematic analysis of the firm productivity distribution.
In a recent paper (Juhász et al. 2020), we examine a unique historical setting that allows us to study technology adoption in the short-run and long-run for the entire firm productivity distribution. We examine the adoption of mechanised cotton spinning in France during the First Industrial Revolution. To do so, we construct a novel firm-level dataset for multiple industries in France from handwritten archival data sources around 1800. This allows us to study the evolution of productivity at the firm level from a remarkably early period during the Industrial Revolution in France. Our findings suggest that the need to reorganise production to make efficient use of new technologies – a feature common to many new technologies – can lead to both slow technology adoption and to aggregate productivity gains that materialise slowly.
Mechanised cotton spinning was invented in Britain in the late 18th century with the development of the famous spinning jenny. It led to huge productivity improvements in the sector. The technology it replaced, handspinning, was conducted by workers in their homes. Mechanised cotton spinning required production to be centrally organised and led to the development of the factory system. This is key, as it allows us to separate production units using the new technology, i.e. all firms, from those operating the old spinning wheels in home production. We are able to follow mechanised cotton spinning firms from the beginning of the widespread use of the spinning jenny in France in the early 1800s, to the period when the technology reached maturity around 1840. Crucially, the technology changed relatively little between 1800 and 1840. Thus, the fact that all firms in cotton spinning operated the new technology allows us to observe productivity differences that are driven by the (more or less efficient) operation of the new technology, as opposed to differences driven by diverse technology vintages across firms, as is common in most datasets.
We compare the evolution of the firm productivity distribution in mechanised cotton spinning to two comparison sectors – metallurgy and paper milling. Production in these two sectors was already organised in firms well before the Industrial Revolution (because of their reliance on high fixed cost machinery and water power). Technological progress in these sectors was more gradual, driven by integrating new vintages of machinery into existing production units – as opposed to the radical shift from home-based to factory-based production in mechanised cotton spinning. This renders these sectors more similar to those observed in typical firm-level datasets.
Our main findings are illustrated in Figure 1, where we plot the distribution of labour productivity across the three sectors, first in 1800 and later, in 1840. Three features stand out. First, the productivity distribution in mechanised cotton spinning was highly dispersed in the earlier (1806) relative to the later period (1840). Second, we estimate that the industry underwent a substantial (82%) increase in productivity after mechanised cotton spinning was adopted (recall that all firms in both 1806 and 1840 used the new technology). Third, and most strikingly, productivity improvements in cotton spinning were driven almost entirely by the disappearance of lower-tail firms. In contrast, in the comparison sectors, the entire firm productivity distribution shifted right.
Figure 1 Changes in the productivity distribution across the three sectors
a) Cotton spinning
c) Paper milling
We argue that the most likely mechanism that explains these findings is the learning via trial and error that was required to efficiently organise the production in cotton mills. Adopting mechanised production methods required entrepreneurs to revolutionise the organisation of production. Within the space of a handful of years, production moved from workers’ homes to large-scale factories, the likes of which had not been seen anywhere else in the economy previously. The historian Stanley Chapman (1974) has described this process as the emergence of the ‘fully-evolved factory’. Our data confirm this: mean firm size (measured as number of employees) was 63 in mechanised cotton spinning in 1806, 20 in metallurgy (firm survey from 1811) and 11 in paper milling (firm survey from 1794). Factory-based production in cotton spinning required entrepreneurs to develop solutions to novel organisational challenges along multiple dimensions. This included designing efficient mill layout and building structures, powering machines effectively, and recruiting and managing a workforce not used to the discipline and division of labour inherent to factory-based production. The knowledge required to solve these challenges had not yet been developed because cotton spinning was the industry to pioneer factory-based production on this scale. Bob Allen (2009, p.184) summarises this most succinctly when he writes, “The cotton mill, in other words, had to be invented as well as the spinning machinery per se.
We present several empirical findings in the paper consistent with a learning mechanism whereby knowledge about efficiently operating the cotton technology diffused only slowly across firms. First, firm survival rates were much lower around 1800 in mechanised cotton spinning relative to the two comparison sectors. This is consistent with early adopters facing considerable uncertainty about best-practice methods in production. In line with this finding, exiting firms in cotton spinning were substantially less productive than those in the comparison sectors. Second, using firm age we show that in the early period (1800), mechanised cotton spinners that entered the market later (i.e. firms that were younger at the time of the survey) had higher productivity. This is consistent with the idea that later entrants had a better pool of knowledge from which to draw best-practice organisational methods. The robust pattern of young firms being more productive is only observed in the early period in cotton spinning and not in 1840, nor in either period in metallurgy where a similar exercise can be conducted. Finally, we provide evidence supporting the spatial diffusion of knowledge: Firms located closer to high-productivity firms were also more productive, and this effect is strong only for cotton spinning firms and only during the initial period of technology adoption.
Our findings suggest that firms may face a strategic incentive to delay adoption in settings where they face high uncertainty about how to organise production to make use of the new technology efficiently. This in turn can lead to a process of slow technology diffusion. Second, our findings speak directly to why aggregate productivity gains from adopting a new technology may take time to materialise. If early adopters need to develop complementary organisational responses to use the technology efficiently, many early users of the technology will operate the technology inefficiently. As a consequence, the full benefits of the new technology materialise relatively slowly for the average firm, and hence take time to show up in aggregate productivity.
Allen, R C (2009), The British Industrial Revolution in Global Perspective, Cambridge University Press.
Atkin, D, A Chaudhry, S Chaudry, A K Khandelwal and E Verhoogen (2017), “Organizational barriers to technology adoption: Evidence from soccer-ball producers in Pakistan”, The Quarterly Journal of Economics 132(3): 1101–1164.
Bandiera, O and I Rasul (2006), “Social networks and technology adoption in Northern Mozambique”, The Economic Journal 116: 869–902.
Beaman, L, J Magruder and J Robinson (2014), “Minding small change among small firms in Kenya”, Journal of Development Economics 108: 69–86.
BenYishay, A and A M Mobarak (2018), “Social learning and incentives for experimentation and communication”, The Review of Economic Studies 86: 976–1009.
Bloom, N, B Eifert, A Mahajan, D McKenzie and J Roberts (2013), “Does management matter? Evidence from India”, Quarterly Journal of Economics 128(1): 1–51.
Chapman, S D (1974), “The textile factory before Arkwright: A typology of factory development”, Business History Review 48(4): 451–478.
Conley, T G and C R Udry (2010), “Learning about a new technology: Pineapple in Ghana”, American Economic Review 100: 35–69.
Duflo, E, M Kremer and J Robinson (2011), “Nudging farmers to use fertilizer: Theory and experimental evidence from Kenya”, American Economic Review 101: 2350–90.
Emerick, K, A de Janvry, E Sadoulet and M H Dar (2016), “Technological innovations, downside risk, and the modernization of agriculture”, American Economic Review 106: 1537–1561.
Foster, A D and M R Rosenzweig (1995), “Learning by doing and learning from others: Human capital and technical change in agriculture”, Journal of Political Economy 103: 1176–1209.
Giorcelli, M (2019), “The Long-Term Effects of Management and Technology Transfers”, American Economic Review 109(1): 1–33.
Hall, B H (2004), “Innovation and diffusion”, NBER Working Paper 10212.
Hall, B H and B Khan (2003), “Adoption of New Technology”, NBER Working Paper 9730.
Hanna, R, M Sendhil and J Schwartzstein (2014), “Learning through noticing: theory and experimental evidence in farming”, Quarterly Journal of Economics 129: 1311–1353.
Hardy, M and J McCasland (2016), “It Takes Two: Experimental Evidence on the Determinants of Technology Diffusion”, Working Paper.
Juhász, R, M P Squicciarini and N Voigtländer (2020), “Technology Adoption and Productivity Growth During the Industrial Revolution: Evidence from France”, NBER Working Paper 27503.
Munshi, K (2004), “Social learning in a heterogeneous population: technology diffusion in the Indian Green Revolution”, Journal of Development Economics 73: 185–213.
Rosenberg, N (1976), “Factors affecting the diffusion of technology”, Explorations in Economic History 10(1): 3–33.
Suri, T (2011), “Selection and comparative advantage in technology adoption”, Econometrica 79: 159–209.
Syverson, C (2011), “What Determines Productivity?”, Journal of Economic Literature 49(2): 326–365.
1 Technology adoption in agriculture has been more extensively studied. See for example Foster and Rosenzweig (1995), Munshi (2004), Bandiera and Rasul (2006), Conley and Udry (2010), Duflo et al. (2011), Suri (2011), Hanna et al. (2014), BenYishay and Mobarak (2018), Beaman et al. (2014), and Emerick et al. (2016).
2 New York Times, 12 July 1987, p. 36