Automatic Reaction – What Happens to Workers at Firms that Automate?

James Bessen, Maarten Goos, Anna Salomons, and Wiljan van den Berge

New AI and robotics technologies are increasingly automating work tasks. How much of a threat does automation pose to workers? A new study provides the first large-scale quantitative evidence of the effects of automation on individual workers, using government data from 2000-2016 for 36,000 firms in the Netherlands, covering about 5 million workers each year.

The study finds that automation does indeed affect many workers. Each year, about 9% of the workers in the sample are employed at firms that make major investments in automation. Yet relatively few workers are adversely affected. Only about 2% of tenured workers at automating firms leave the year of the automation event as a result of automation; after 5 years, 8.5% will have left, cumulatively. Nevertheless, those who do leave suffer significant economic costs largely due to unemployment spells. Government safety net programs do not nearly compensate these workers for their losses. Surprisingly, this burden falls more frequently on highly educated and highly paid workers, contrary to conventional wisdom; they are more likely to leave as a result of automation, although they also seem to find new jobs faster.

To understand what happens when firms introduce automation on a large scale, the paper examines “automation spikes,” which it defines as a year in which a firm makes expenditures on automation that are at least three times its average spending on automation in all other years.  Comparing workers who experience such a spike in a given year with a control group that experiences spikes later enables analysis of how the spikes affect workers. The paper looks at both tenured workers (3 years or more at the firm) and recently hired workers.

Although many commentators liken the introduction of automation in a workplace to a mass layoff or a plant closing, this study shows that that comparison is not particularly apt and those fears are overblown.  The data show that workers do experience loss of both earnings and work following a spike, but that loss is substantially less than that experienced by workers following a mass layoff. In the sample, only 0.7% of all workers on average leave their employers each year due to automation.  In contrast, in the Netherlands, about  3.5%-7.2% lose their jobs each year in mass layoffs.  (The comparable rate is 4.4% in the U.S.) The risk of anyone losing a job due to automation is thus much smaller than the risk of a mass layoff.  Moreover, while plant closings or mass layoffs affect a large number of workers all at once, the effects from automation happen more gradually, giving workers more time to react and adjust.  

The real impact of automation is on income and time spent unemployed.   The data show that after a spike, tenured workers cumulatively lose about 3,800 Euros in wage and salary earnings over five years on average (about 9% of one year’s income).  More recent hires also experience a negative impact, but only about 3% of one year’s income, which perhaps reflects their adaptability and flexibility as newer hires. The main reason for these losses is that workers who leave experience longer periods of non-employment. Conditional on leaving, incumbent workers are employed 43 fewer days over five years following the automation event. Wage rates change little for both workers who stay at the firm and those who leave and are re-employed elsewhere. Only about 12% of these losses are made up by unemployment, welfare, or disability payments, which is comparable to what workers receive after a mass layoff.

The study also looks at differences across firms and workers with different characteristics. Osborne and Frey (2017) contend that low wage occupations will be hardest hit by automation. Our data reject this common assumption, but we do find that lower paid workers suffer longer unemployment after leaving. Among tenured workers, the probability of leaving employment does not vary much by age or gender, but among recent hires, older workers are more likely to leave. Also, while the losses are more severe in manufacturing industries, the impacts are seen across all industry sectors studied.

The study paints a picture of automation today that does not support the most alarmist views. The burden that automation places on workers is less than the burden created by mass layoffs and plant closings that arise from things like declining demand or bankruptcies. Nevertheless, the burden placed on affected workers is substantial, and existing safety net programs are not providing these workers much economic security. And, of course, the impact of automation might worsen in the future.  Further research will show what happens to net employment after automation, and to the workers hired after the automation event.

SSRN Working Paper