The sometimes puzzling differences in transatlantic earnings growth

The earnings of low-skill workers in the United States have typically grown more slowly than those of high-skill workers. Earnings growth for workers with mid-level skills has typically fallen between the rates for those with high and low skills. Data on earnings by skill level is available from the Atlanta Fed's Wage Growth Tracker starting in 1997 (Figure 1). The dataset shows a similar picture for earnings growth by education: highly-educated workers (who typically have high-level skills) have benefitted from faster wage growth than those with middle levels of education, while the least educated (who typically have low skills) saw the slowest wage growth.

Other research has confirmed that these developments have characterised the US labour market from the early 1980s. In 2022, low-skill and mid-skill wage growth has accelerated notably (Figure 1), but this hardly compensates for the preceding four decades.

But most of Europe is different. A few European countries look like the US in terms of earnings growth, but in many, wages of workers in low-skill occupations increased faster than wages of high-skill occupations from 2006 to 2018, as we demonstrated in a paper for the Transatlantic Expert Group on the Future of Work.

It should be noted that European wage data is scarcer than data for the US. Eurostat’s frequently updated quarterly labour cost indicators do not differentiate according to individual characteristics of employees, like occupation and educational level. For such data, the Structure of Earnings Survey provides information every fourth year. The latest available data is for 2018, while the first observation is 2006 for most EU countries.

Figure 2 reports data for six of the ten main occupational categories: managers and professionals are usually classified as having the highest skills and their earnings are higher than in other occupations; technicians and clerical support workers typically have mid-level qualifications and are in the middle of the income distribution; workers in sales and elementary occupations typically have low-level qualifications, low skills and low earnings.

Among 26 EU countries (data for Croatia is not available from 2006), the high-skilled benefitted from the fastest earnings growth only in Bulgaria, Ireland and the Netherlands, with managers and professionals generally seeing the fastest earnings growth. In Malta, the two mid-skill occupations (technicians and clerical workers) saw the fastest earnings growth, followed by the two high-skill occupations, while low-skill occupations (sales and elementary occupations) saw the slowest earnings growth.

In eleven countries, meanwhile, low-skill earners experienced the fastest earnings growth (Austria, Estonia, Hungary, Latvia, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, and Sweden): the two low-skill occupations were either the top two or in the top three for earnings growth, outpacing both high-skill occupational categories.

In the remaining eleven countries the evidence is mixed, yet there is at least one low-skill occupation that benefitted from faster earnings growth than at least one high-skill occupation. The same mixed outcome characterises the United Kingdom.

In the US, automation has driven wage inequality

Explanations for diverging earnings growth depending on skills and educational levels include technological change, globalisation and changes in labour markets, such as the declining bargaining power of low-skill workers or the relative reduction in minimum wages. However, technological change may have played the dominant role in the United States.

For example, Daron Acemoglu and Pascual Restrepo argued in 2021 that the relative wage declines experienced by workers doing routine tasks in industries experiencing rapid automation explain between 50% and 70% of changes to the US wage structure over the last four decades. Examples of automation include industrial robots replacing blue-collar workers in manufacturing and specialised software replacing clerical workers. They also concluded that offshoring contributes to task displacement, but accounts for a smaller portion of observed wage changes than automation. Rising market power, markups, import competition and deunionisation do not appear to play major roles in US wage inequality.

Some compelling theories suggest that automation can be the main driver of wage inequality. Acemoglu and Restrepo also developed a framework in which automation technologies expand the set of tasks performed by capital, and displace workers previously employed in these tasks. Workers displaced from routine tasks typically have low or mid-level skills. Displaced workers compete against others for non-automated tasks, bid down their wages and spread the negative wage effects of automation more broadly through the population. The positive impact of task displacement on productivity, GDP and the average wage level could be a counter-balancing factor, but this effect was found modest.

David Autor in 2019 also argued that technological change has played a major role in US wage inequality and highlighted the issue of polarisation: the hollowing out of mid-skill, non-college blue-collar production and white-collar administrative support jobs. He suggested three mechanisms to explain why. First, non-college workers have been shunted from mid-skill career occupations that reward specialised and differentiated skills, into traditionally low-education occupations that demand generic skills. Second, technological change has depressed mid-wage employment disproportionately among non-college workers in urban areas, thus directly reducing average non-college wages. Third, technology has created an excess supply of less-educated workers that depresses non-college wages generally.

Other theories show that if high-skilled workers are complements to machines and low-skilled workers are substitutes for machines, then automation leads to an increasing share of college graduates in the population, increasing income and wealth inequality, and causes a declining labour share in total income. Greater complementarity between robots and older workers, and greater substitutability between robots and younger workers, has also been found. Since older workers tend to earn more than younger workers, this process also increases inequality. Automation could also increase inequality via increased returns to wealth.

In Europe, the role of automation in inequality is unclear

European labour markets are similar in some ways to US labour markets, such as in the increase in jobs with high-level education attainment (and also an increase in people with tertiary education) and the reduction in jobs with low-levels of educational attainment. Polarisation is also observed in Europe: the numbers of high-skill jobs and the numbers of low-skill service occupations have grown. By contrast, the number of middle-skill jobs (such as clerks, machine operators and assemblers) has declined in Europe.

Polarisation in Europe is illustrated by changing employment shares. From 2002 to 2016, the employment share of high-skill workers increased by 8 percentage points, that of low-skilled workers increased by 1 percentage point, while that of mid-skilled workers fell by 9 percentage points. Low-skill and mid-skill jobs are significantly exposed to automation.

In terms of the extent of automation, the US and the EU saw similar trends from 1993 to 2016 in the deployment of industrial robots, while robot density (number of robots per number of workers) has consistently been higher in the EU than in the US. The EU lags behind the US in artificial intelligence investment (by a large margin) and adoption (by a smaller margin). The World Economic Forum’s ITC adoption rankings puts ten, mostly small, EU countries ahead of the US, and five small EU countries ahead in terms of digital skills. The EU’s Digital Economy and Society Index shows big differences across the EU, but it does not compare to the US.

Thus, there are several similarities between the EU and US in labour-market and technology-adoption trends, though the US is more advanced in digital technologies. It’s puzzling that, if automation is the main driver of US earnings inequality, why similar earnings growth inequality is not seen in Europe.

It may be that labour market characteristics influence how the gains from automation are shared more with workers. One analysis found no overall effect of robotisation on average earnings in 20 European countries from 2010 to 2015, but found a positive effect on average earnings for workers in countries with above-median collective bargaining coverage. Collective bargaining might reduce inequality. Other ‘European’ factors could include labour-market regulations, minimum wages and differences in industrial structure and technology adoption. Further research should study the role of these and other possible factors in explaining transatlantic differences – and within-EU differences – in earnings growth.