/The Individual Costs of Occupational Decline

The Individual Costs of Occupational Decline

Yves here. You have to read a bit into this article on occupational decline, aka, “What happens to me after the robots take my job?” to realize that the authors studied Swedish workers. One has to think that the findings would be more pronounced in the US, due both to pronounced regional and urban/rural variations, as well as the weakness of social institutions in the US. While there may be small cities in Sweden that have been hit hard by the decline of a key employer, I don’t have the impression that Sweden has areas that have suffered the way our Rust Belt has. Similarly, in the US, a significant amount of hiring starts with resume reviews with the job requirements overspecified because the employer intends to hire someone who has done the same job somewhere else and hence needs no training (which in practice is an illusion; how companies do things is always idiosyncratic and new hires face a learning curve). On top of that, many positions are filled via personal networks, not formal recruiting. Some studies have concluded that having a large network of weak ties is more helpful in landing a new post than fewer close connections. It’s easier to know a lot of people casually in a society with strong community institutions.

The article does not provide much in the way of remedies; it hints at “let them eat training” when programs have proven to be ineffective. One approach would be aggressive enforcement of laws against age discrimination. And even though some readers dislike a Job Guarantee, not only would it enable people who wanted to work to keep working, but private sector employers are particularly loath to employ someone who has been out of work for more than six months, so a Job Guarantee post would also help keep someone who’d lost a job from looking like damaged goods.

By Per-Anders Edin, Professor of Industrial Relations, Uppsala University; Tiernan Evans, Economics MRes/PhD Candidate, LSE; Georg Graetz, Assistant Professor in the Department of Economics, Uppsala University; Sofia Hernnäs, PhD student, Department of Economics, Uppsala University; Guy Michaels,Associate Professor in the Department of Economics, LSE. Originally published at VoxEU

As new technologies replace human labour in a growing number of tasks, employment in some occupations invariably falls. This column compares outcomes for similar workers in similar occupations over 28 years to explore the consequences of large declines in occupational employment for workers’ careers. While mean losses in earnings and employment for those initially working in occupations that later declined are relatively moderate, low-earners lose significantly more.

How costly is it for workers when demand for their occupation declines? As new technologies replace human labour in a growing number of tasks, employment in some occupations invariably falls. Until recently, technological change mostly automated routine production and clerical work (Autor et al. 2003). But machines’ capabilities are expanding, as recent developments include self-driving vehicles and software that outperforms professionals in some tasks. Debates on the labour market implications of these new technologies are ongoing (e.g. Brynjolfsson and McAfee 2014, Acemoglu and Restrepo 2018). But in these debates, it is important to ask not only “Will robots take my job?”, but also “What would happen to my career if robots took my job?”

Much is at stake. Occupational decline may hurt workers and their families, and may also have broader consequences for economic inequality, education, taxation, and redistribution. If it exacerbates differences in outcomes between economic winners and losers, populist forces may gain further momentum (Dal Bo et al. 2019).

In a new paper (Edin et al. 2019) we explore the consequences of large declines in occupational employment for workers’ careers. We assemble a dataset with forecasts of occupational employment changes that allow us to identify unanticipated declines, population-level administrative data spanning several decades, and a highly detailed occupational classification. These data allow us to compare outcomes for similar workers who perform similar tasks and have similar expectations of future occupational employment trajectories, but experience different actual occupational changes.

Our approach is distinct from previous work that contrasts career outcomes of routine and non-routine workers (e.g. Cortes 2016), since we compare workers who perform similar tasks and whose careers would likely have followed similar paths were it not for occupational decline. Our work is also distinct from studies of mass layoffs (e.g. Jacobson et al. 1993), since workers who experience occupational decline may take action before losing their jobs.

In our analysis, we follow individual workers’ careers for almost 30 years, and we find that workers in declining occupations lose on average 2-5% of cumulative earnings, compared to other similar workers. Workers with low initial earnings (relative to others in their occupations) lose more – about 8-11% of mean cumulative earnings. These earnings losses reflect both lost years of employment and lower earnings conditional on employment; some of the employment losses are due to increased time spent in unemployment and retraining, and low earners spend more time in both unemployment and retraining.

Estimating the Consequences of Occupational Decline

We begin by assembling data from the Occupational Outlook Handbooks (OOH), published by the US Bureau of Labor Statistics, which cover more than 400 occupations. In our main analysis we define occupations as declining if their employment fell by at least 25% from 1984-2016, although we show that our results are robust to using other cutoffs. The OOH also provides information on technological change affecting each occupation, and forecasts of employment over time. Using these data, we can separate technologically driven declines, and also unanticipated declines. Occupations that declined include typesetters, drafters, proof readers, and various machine operators.

We then match the OOH data to detailed Swedish occupations. This allows us to study the consequences of occupational decline for workers who, in 1985, worked in occupations that declined over the subsequent decades. We verify that occupations that declined in the US also declined in Sweden, and that the employment forecasts that the BLS made for the US have predictive power for employment changes in Sweden.

Detailed administrative micro-data, which cover all Swedish workers, allow us to address two potential concerns for identifying the consequences of occupational decline: that workers in declining occupations may have differed from other workers, and that declining occupations may have differed even in absence of occupational decline. To address the first concern, about individual sorting, we control for gender, age, education, and location, as well as 1985 earnings. Once we control for these characteristics, we find that workers in declining occupations were no different from others in terms of their cognitive and non-cognitive test scores and their parents’ schooling and earnings. To address the second concern, about occupational differences, we control for occupational earnings profiles (calculated using the 1985 data), the BLS forecasts, and other occupational and industry characteristics.

Assessing the losses and how their incidence varied

We find that prime age workers (those aged 25-36 in 1985) who were exposed to occupational decline lost about 2-6 months of employment over 28 years, compared to similar workers whose occupations did not decline. The higher end of the range refers to our comparison between similar workers, while the lower end of the range compares similar workers in similar occupations. The employment loss corresponds to around 1-2% of mean cumulative employment. The corresponding earnings losses were larger, and amounted to around 2-5% of mean cumulative earnings. These mean losses may seem moderate given the large occupational declines, but the average outcomes do not tell the full story. The bottom third of earners in each occupation fared worse, losing around 8-11% of mean earnings when their occupations declined.

The earnings and employment losses that we document reflect increased time spent in unemployment and government-sponsored retraining – more so for workers with low initial earnings. We also find that older workers who faced occupational decline retired a little earlier.

We also find that workers in occupations that declined after 1985 were less likely to remain in their starting occupation. It is quite likely that this reduced supply to declining occupations contributed to mitigating the losses of the workers that remained there.

We show that our main findings are essentially unchanged when we restrict our analysis to technology-related occupational declines.

Further, our finding that mean earnings and employment losses from occupational decline are small is not unique to Sweden. We find similar results using a smaller panel dataset on US workers, using the National Longitudinal Survey of Youth 1979.

Theoretical implications

Our paper also considers the implications of our findings for Roy’s (1951) model, which is a workhorse model for labour economists. We show that the frictionless Roy model predicts that losses are increasing in initial occupational earnings rank, under a wide variety of assumptions about the skill distribution. This prediction is inconsistent with our finding that the largest earnings losses from occupational decline are incurred by those who earned the least. To reconcile our findings, we add frictions to the model: we assume that workers who earn little in one occupation incur larger time costs searching for jobs or retraining if they try to move occupations. This extension of the model, especially when coupled with the addition of involuntary job displacement, allows us to reconcile several of our empirical findings.


There is a vivid academic and public debate on whether we should fear the takeover of human jobs by machines. New technologies may replace not only factory and office workers but also drivers and some professional occupations. Our paper compares similar workers in similar occupations over 28 years. We show that although mean losses in earnings and employment for those initially working in occupations that later declined are relatively moderate (2-5% of earnings and 1-2% of employment), low-earners lose significantly more.

The losses that we find from occupational decline are smaller than those suffered by workers who experience mass layoffs, as reported in the existing literature. Because the occupational decline we study took years or even decades, its costs for individual workers were likely mitigated through retirements, reduced entry into declining occupations, and increased job-to-job exits to other occupations. Compared to large, sudden shocks, such as plant closures, the decline we study may also have a less pronounced impact on local economies.

While the losses we find are on average moderate, there are several reasons why future occupational decline may have adverse impacts. First, while we study unanticipated declines, the declines were nevertheless fairly gradual. Costs may be larger for sudden shocks following, for example, a quick evolution of machine learning. Second, the occupational decline that we study mainly affected low- and middle-skilled occupations, which require less human capital investment than those that may be impacted in the future. Finally, and perhaps most importantly, our findings show that low-earning individuals are already suffering considerable (pre-tax) earnings losses, even in Sweden, where institutions are geared towards mitigating those losses and facilitating occupational transitions. Helping these workers stay productive when they face occupational decline remains an important challenge for governments.

Please see original post for references

Print Friendly, PDF & Email