American jobs are starting to look more alike. An analysis of hundreds of millions of US job ads posted between 2014 and 2022 finds that the skills employers ask for in different occupations have been steadily converging, and that this convergence goes hand in hand with narrowing pay gaps between occupations and more workers switching between them.
The work comes from Alexis Antoniades and Emmanouil Chatzikonstantinou of Georgetown University in Qatar and Mohammed Shahmeer Ahmad of the Qatar Computational Research Institute. Their working paper, “The Convergence of Occupations: Evidence from Online Job Posts,” is dated December 16, 2024 and is available on SSRN.
A reversal of an old trend
For decades, technology was thought to pull occupations apart. Skill-biased technical change favoured some workers over others and drove rising inequality and falling mobility from the 1980s on.
The authors point to a break. Using US wage data, they show that pay differences across occupations widened until 2008, flattened, and have been shrinking since 2014. Mobility between occupations, measured in the Current Population Survey, fell until 2010 and has risen since.
Their explanation is that new technology now transforms occupations in a way that brings them closer together. Old skills such as typewriting or teller transactions belonged to a single job and are disappearing. New skills such as data analytics and Python spread across many jobs at once.
As the authors put it, a data scientist “can transition seamlessly from research to marketing, agriculture, or banking,” while a teller is confined to banking.
Measuring the distance between 1,326 jobs
The data come from Lightcast, formerly Burning Glass Technologies, which scrapes job postings daily from 40,000 to 50,000 sources and tags each one with the skills it mentions. The dataset tracked about 14,000 distinct skills.
The researchers turned each of 1,326 detailed occupations into a weighted list of the skills its ads request. They then calculated the distance between every pair of occupations, year by year, using first-quarter data from 2014 to 2022.
The shift in demand for new skills was large. Across all ads, the share of skill mentions tied to artificial intelligence rose from 0.5% in 2014 to 4.5% in 2022, and the share tied to data analytics rose from 2.3% to 8.5%.
Both the average and the median distance between occupations fell every year in that window. The result held when the team swapped their main distance measure for an alternative one. Every broad occupational group moved closer to the others, with military occupations converging the most and social services the least.
Closer skills, closer pay, more switching
The authors then tested whether pairs of occupations that grew more similar also saw their wages converge. They did. The bigger the initial pay gap in 2014, the more it narrowed, and pairs whose skill profiles moved closer showed significantly more wage convergence even after accounting for that starting gap.
Occupations low in non-routine cognitive work saw the largest drop in wage inequality between 2014 and 2022. The authors note that this reverses the pre-2010 pattern, when inequality was rising for exactly those jobs.
Mobility followed the same logic. Pairs of occupations whose skill distance shrank saw significantly larger increases in the share of workers moving from one to the other, a result that survived controls for occupation fixed effects and weighting by employment.
Occupations that changed the most internally, measured as their distance from their own 2014 profile, also showed higher wage growth and larger falls in pay inequality within the occupation.
The paper also maps the job market as a network. Business and management specialists sit at its centre, the occupations most closely linked to everything else.
Study Details:
- Title: The Convergence of Occupations: Evidence from Online Job Posts
- Authors: Alexis Antoniades, Emmanouil Chatzikonstantinou, Mohammed Shahmeer Ahmad
- Journal: SSRN working paper
- Publication Date: December 16, 2024
- DOI: 10.2139/ssrn.5096487