VoxEU Column Labour Markets Productivity and Innovation

The labour market consequences of technology adoption: Concrete evidence from the Great Depression

A longstanding question in economics is whether labour-saving technology affects firms in the medium term by increasing output, by decreasing employment, or both. This column provides evidence on this issue using a novel dataset from the concrete industry during the Great Depression. Cheaper electricity caused a decrease in the labour share of income, an increase in productivity and electrical capital intensity, and a decrease in employment. Furthermore, these effects were stronger in counties where the Depression hit hardest, consistent with the idea of ‘the cleansing effect of recessions’.

Do labour-saving technologies lead to a fall in employment or an increase in output in the medium run? This debate on labour displacement seems to be revived with each new wave of technology adoption – from the Jacquard textile loom, to the steam engine, to electricity, to computers.

On the one hand, proponents argue that either firms do not pass on lower production costs as lower prices, or that demand adjusts slowly to lower output prices. In this scenario, labour-saving innovations improve labour productivity faster than demand for products and are bound to displace some types of workers in the medium term, i.e. over a 5-10 year horizon (Ricardo 1821, Mill 1871, Keynes 1933, Samuelson 1988). This is the view of ‘technological unemployment’.

On the other hand, critics argue that firms do pass on lower production costs as cheaper prices and that demand for products adjusts quickly to those prices. In this scenario, the effect of productivity-enhancing technologies should occur at the output margin with firms increasing production instead of destroying jobs (McCulloch 1821, Say 1924, Woirol 1996). Opponents refer to this view as the ‘Luddite fallacy’.

The debate has again resurfaced in the context of information technologies, e.g. with computers leading firms to create more narrow job opportunities by skill and permanently increasing unemployment and the skill premium (Acemoglu 1999), and especially with the high unemployment since the Great Recession of 2007 (Jaimovich and Siu 2012, Krugman 2013, Frey and Osborne 2013). Despite the length of the debate and the relevance of occupational displacement for policymakers since the financial crisis of 2007, there is little empirical work supporting either side of the discussion.

New research on the labour market consequences of technology adoption

In a recent paper (Morin 2015), I exploit the within-country variation in the cost of electric technology, which is crucial for US in the early 20th century. My approach improves on the existing literature in two respects.

First, I identify the causal effects of labour-saving technology using a novel identification strategy. Among the three important General Purpose Technologies of steam, electricity, and computers, it is electricity that has the most within-country variation in price. Furthermore, a regression of quantities on prices raises concerns about endogeneity and is a challenge to identification – it is unclear whether the regression estimates the demand or supply equation.

An instrument that shifts the electricity supply curve is the coal share of power. Regions with hydroelectric power had initially cheaper electricity but the price fell slowly because hydroelectric technology was very efficient since the beginning and had little margin to improve. In contrast, regions with coal power had initially more expensive electricity but the price fell much faster because coal technology was improving steadily. The electricity market was regulated at the state-level with limited cross-state transfers, making the state the relevant level of variation for prices.

Figure 1 illustrates the predictive power of the coal share for the fall in the price of electricity at the state-level.

Figure 1 Initially higher coal share causes a decrease in the price of electricity.

Note that the instrument (coal share in the x-axis) and endogenous variable (fall in price of electricity in the y-axis) vary at the state-level, hence the 42 observations. The dependent variables (productivity, labour share, employment) vary at the level of concrete plants, which are represented by the size of the circles in each state.

Second, I focus on the concrete industry—a local, non-traded industry whose products harden quickly and weigh a considerable amount, precluding their long-distance transport. This non-traded aspect means that concrete firms locate near their customers, not near cheap electricity, and implies that I am able to focus on the economic mechanism at work without worrying about the confounding strategic decisions of firms’ locations.

The novel dataset was collected and digitised from thousands of handwritten archival sources. It provides detailed measurements at the level of each concrete plant for the labour share of income, labour productivity, electric capital intensity (electric horsepower per worker), output, and employment.

In the absence of within-country variation in the price of a technology, existing studies have had to use anecdotal evidence or aggregate data with potentially confounding events. As such, this study constitutes the first credibly identified evidence on the labour market consequences of technology adoption.

Empirical findings

I find that electricity was a labour-saving technology and that firms adjusted by firing workers instead of increasing output. Cheaper electricity caused an increase in electric capital intensity and in labour productivity, as well as a decrease in the labour share of income. It explains between 15% and 25% of the change in these variables.

I also find evidence for the technological unemployment hypothesis: cheaper electricity had no statistically significant effect on output, either in quantity or value, and explains 15% of the decrease in employment. To put this in perspective, electricity is a better predictor of employment change than the boom in housing during the 1920s.

These results are consistent with the view that the adoption of labour-saving technology causes job loss in the adopting sector. Other mechanisms occurred simultaneously and the results are robust to these potentially confounding mechanisms – the importance of agriculture, the housing boom in the 1920s, proximity to dam construction, and entry and exit of firms.

I also find that these effects were stronger in counties where the Depression hit harder, as measured by the change in farm output at the county-level between 1930 and 1935 (farming represented 40% of employment and these results are similar when using the housing bust as a measure of the severity of the depression). The causal effect of cheaper electricity is larger for the labour share of income, labour productivity and employment in counties where the Depression was more severe.

These results are consistent with the ‘cleansing effect of recessions’ where a temporary downturn forces firms to adopt labour-saving technologies that increase productivity and reduce labour input (Caballero and Hammour 1994, Field 2003).

Figure 2 Causal effect of electricity on labour market outcomes for counties with a by deep depression (red) and shallow depression (green).

The regression is an instrumental variable regression of an outcome on electricity prices instrumented by the coal share. The sample is split into above- and below-median change in agricultural output. The whiskers represent 95% confidence intervals. First-stage F-statistics and number of states are at the right

I also estimate a structural equation for the labour share of income to pin down the elasticity of substitution between electric capital and dexterity tasks. This estimation is done in OLS and without the coal share as an instrument (the non-linear equation for the labour share requires using the level of electricity prices at the beginning and end of the period and requires two instruments, whereas the paper has only one).

The estimate of σ = 2.2 is above 1, like Karabarbounis and Neiman (2012) and unlike Oberfield and Raval (2014). It is within a third of the standard error of the estimate of Krusell et al. (2000) for the elasticity of substitution between capital equipment and unskilled labour for the post-war period.

Concluding remarks

My findings lend support to the theory of technological unemployment – firms reacted to cheaper electricity by reducing the price of concrete products, but the demand for concrete was not sufficiently elastic and the growth in output was too weak to offset productivity gains. They are also consistent with the ‘cleansing effect of recessions’, where temporary downturns allow firms to take advantage of new labour-saving technologies, restructure production, and reduce labour input.

Two questions naturally arise after this work.

  • First, do these results generalise from the concrete industry to the whole economy?

Counter-arguments for this generalisation include a strand of the labour-saving technologies debate, whereby consumers benefit from cheaper concrete prices and expand demand in other industries.

Supporting arguments for this generalisation are the broad representativeness of concrete for the whole economy (it is in the median of the distribution of electricity share of value added in levels and in changes) and the clarity of the results for this industry.

  • Second, does electricity in the 1930s generalise to technologies today?

This is also an open question and supporting arguments are the similarities between computers and electricity (David 1990) and the work of other authors who have linked labour market changes since the 1980s to computers, e.g. Jaimovich and Siu (2012).


Acemoglu, D (1999) “Changes in Unemployment and Wage Inequality: An Alternative Theory and Some Evidence”, The American Economic Review 89(5):1259–1278, December.

Caballero, R and M Hammour (1994), “The cleansing effect of recessions”, The American Economic Review 84(5):1350–1368.

David, P (1990) “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox”, The American Economic Review 80(2):355–361, May.

Field, A (2003) “The Most Technologically Progressive Decade of the Century”, The American Economic Review 93(4):1399–1413, September.

Benedikt, F and Osborne, M (2013), “The future of employment: how susceptible are jobs to computerisation?”, University of Oxford Manuscript, access October.

Jaimovich, N and H Siu (2012), “The Trend is the Cycle: Job Polarisation and Jobless Recoveries”, NBER Working Paper 18334.

Karabarbounis, L and B Neiman (2012), “Declining Labour Shares and the Global Rise of Corporate Savings”, NBER Working Paper 18154.

Keynes, J M (1933), “Economic possibilities for our grandchildren”, Essays in Persuasion, Macmillan.

Krugman, P (2013), “Sympathy for the Luddites”, The New York Times, June.

Krusell, P, L Ohanian, J-V Rios-Rull and G Violante (2000), “Capital-Skill Complementarity and Inequality: A Macroeconomic Analysis”, Econometrica 68(5):1029–1053.

McCulloch, J R (1821), “The Opinions of Messrs. Say, Sismondi and Malthus on the Effects of Machinery and Accumulation, Stated and Examined”, The Edinburgh Review, March.

Mill, J S (1871), Principles of Political Economy with Some of their Applications to Social Philosophy, Longmans, Green, Reader and Dyer, London.

Morin, M (2015), “The labour market consequences of electricity adoption: Concrete evidence from the great depression”, INET Working Paper 1511.

Oberfield, E and D Raval (2014), “Micro data and macro technology”, NBER working paper 20452.

Ricardo, D (1821), On The Principles of Political Economy and Taxation, John Murray, London, 3rd edition.

Samuelson, P (1988), “Mathematical Vindication of Ricardo on Machinery”, Journal of Political Economy 96(2):274–282, April.

Say, J-B (1924), Treatise on Political Economy: Or the Production, Distribution, and Consumption of Wealth, volume I, Boston: Wells and Lilly, 2nd American edition.

Woirol, G (1996), The Technological Unemployment and Structural Unemployment Debates, Greenwood Press, Westport, Connecticut.

9,980 Reads