Automation and Employment
What happens when firms make investments in automation? While this is about firm strategy, the answer to the question is of the utmost importance to public policy, namely public policy regarding innovation and employment.
Over time, we observe firms making investments in automation and we observe employment trends. For a researcher, the challenge is to distinguish between correlation and causality: are the observed trends in employment caused by the observed trends in automation, or is it simply a case of correlation? The question is particularly relevant to the extent that automating and non automating firms are on different employment trends.
A recent paper attacks this problem by developing “a novel empirical approach based on evidence that firms’ automation expenditures occur in discrete episodes of lumpy investment.” In other words, the authors make use of spikes in innovation rates. The data, provided by Statistics Netherlands, covers the 2000–2016 period and corresponds to 35,580 unique firms with at least 3 years of automation cost data. Together, these firms employ around 5 million unique workers annually on average.
The following figure summarizes the data relating to automation spikes. The horizontal axis measures years relative to the automation event; the vertical axis measures the share of automation costs over total cost.
As can be seen, the share of automation cost on total cost is significantly higher at the time of investment. Together with detailed data regarding workers’ employment and wages, the authors are able to track the estimated effects of automation events.
A striking result is the contrast between small and large firms. The figure below presents results of a difference-in-differences analysis of the effect of automation on employment levels and wage levels. The horizontal axis measures years relative to the automation event. In the top panel, the vertical axis corresponds to the log of employment change, so the value corresponds approximately to percent change in employment. For firms with more than 500 workers, the drop is small and in fact not significantly different from zero (the top of the 95% confidence interval lies above zero). By contrast, for firms with less than 500 workers, the estimated effect of automation is significant (about 20% after four years) and statistically significant.
When it comes to wages, the effects are also quite different across firm size: essentially no change for small firms but a 5% increase (approximately) for large firms.
Regarding the effect on workers, the authors estimate a five-year cumulative wage income loss of 9% of one year’s earnings, driven by decreases in days worked. They also find that these adverse impacts of automation correspond to older and middle-educated workers.
By contrast with the above results, no losses are found for firms’ investments in computers.
James Bessen, Maarten Goos, Anna Salomons, Wiljan van den Berge
What Happens to Workers at Firms that Automate?
The Review of Economics and Statistics 107 (2025), 125–141.