« Back to Results

Technological Change and the Value of Skills

Paper Session

Saturday, Jan. 4, 2025 8:00 AM - 10:00 AM (PST)

Hilton San Francisco Union Square, Yosemite A
Hosted By: American Economic Association
  • Chairs:
    Christina Langer, Stanford University
  • Sebastian Steffen, Boston College

Returns to Skills and Apprenticeship Reforms

Christina Langer
,
Stanford University

Abstract

The main challenges when estimating returns to skills are measurement and selection. I exploit quasi-experimental variation stemming from apprenticeship plan modernizations that can be seen as a relatively exogenous shock to address these concerns. I eliminate measurement error by observing the complete picture of skills imparted in an apprenticeship before and after an update. Within occupational variation addresses selection. I find that updates alter the skill landscape of apprenticeships, increasing the relative importance of social, digital, management, and administrative skills to the detriment of cognitive and manual skills. Cognitive, social, and digital skill increases within occupation are further significantly related to higher wages over workers' life cycle.

Complement or Substitute? How AI Increases the Demand for Human Skills

Fabian Stephany
,
Oxford Internet Institute
Elina Mäkelä
,
Oxford Internet Institute

Abstract

The question of whether the effects of AI are to substitute or complement human work is central to debates on the future of work. This paper examines the impacts of artificial intelligence (AI) on the demand and compensation for skills in the U.S. economy. This research marks a contribution to the field in its investigation of both internal (within-job substitution and complementation) and external (substitution and complementation across occupations, industries and regions) effects, utilising a large dataset of 12 million online job vacancies from 2018 to 2023. Our analysis reveals a statistically significant increase in demand for skills identified as complementary to AI technologies (digital literacy, team work, or resilience and agility), alongside a rising premium for these skills within AI roles such as Data Scientist. Conversely, skills that are considered substitutes for AI, such as customer service, summarisation or text review, have declined both in popularity and value within AI-related positions. Expanding our analysis to measure the external effects across the broader economy, we find a significant increase in demand for complementary skills outside of AI roles but which can be linked to the growth of AI roles within those occupations, industries, or regions. In parallel, there is a moderate decline in the demand for non-AI roles that involve substitute skills across occupations, industries and regions that can be linked to AI. Finally, we estimate that AI has led to a net positive demand, since the complementary effect is up to 50% larger than the substitution effect. We can replicate our results for the UK and Australia. These findings underscore the profound and widespread impact of AI on the evolving skill requirements in the workforce, and suggest reskilling efforts not just prioritise technical AI skills but also AI-complementary skills such as ethics or digital literacy.

Superstars at Work: Increasing Returns to Scale Across Occupations

Ruyu Chen
,
Stanford University
Erik Brynjolfsson
,
Stanford University
Seth Benzell
,
Chapman University

Abstract

Labor incomes in the U.S. are characterized by fat tails. The “superstar” theory of wage inequality suggests that this is because information technologies can make certain labor markets “winner-take-all”. These technologies allow workers to better leverage their expertise and market their services across larger markets with fewer frictions, amplifying advantages at the top of the skill distribution. This paper explores how the wages of different occupations vary with market size and IT intensity, to assess whether the superstar labor theory can explain patterns in US wage inequality. Using a large-scale, high-frequency administrative payroll dataset on the employment of over 25 million U.S. workers from 2016 through 2022, we measure how wages scale with establishment, firm and market size across occupations, industries, and time. We replicate the findings of Gabaix and Landier (2008), of CEO wage scaling with firm size on our much larger dataset. We find that most other occupations also exhibit wage scaling. We find wage scaling exists for workers at all wage percentiles for occupations in an occupation-year but is highest for the top-paid worker in an occupation-establishment-year. We find large variations in how wages scale with establishment size. The average job in our data is 16.4% higher paid in an establishment which is twice as large. Greatest scaling is found for occupations that are intensive in leadership and judgment, and low in routine and manual tasks (e.g. managers). We also find that IT hiring at the establishment level and AI exposure at the industry level is associated with stronger superstar wage scaling with firm payroll. This is consistent with the theory that IT can increase span-of-control, giving an increasing share of the labor market winner-take-all attributes.

Technology in the Banking Sector and the Importance of Modern Skills

Anastassia Fedyk
,
University of California-Berkeley
Efraim Benmelech
,
Northwestern University
James Hodson
,
AI for Good Foundation
Vladimir Mukharlyamov
,
Georgetown University
Dimitris Papanikolaou
,
Northwestern University

Abstract

This paper examines technology adoption and individual career progression of technically-skilled employees in the banking sector. We leverage detailed employer-employee data on individual worker skills to capture (i) the technical skills of each employee, (ii) rm-level technical capital, and (iii) a novel methodology to systematically measure vintages of technical skills, differentiating older technical skills (e.g., Fortran) from newer technical skills (e.g., Python). Banks dramatically increase their technology investments during the 2010s, with an average bank's technical capital growing by 84%. This increase is ubiquitous across bank size, deposits, and loan portfolio composition. At the individual level, we document a significant difference between employees with modern technical skills and those with outdated technical skills. Having modern technical skills in 2015 decreases the likelihood of adverse career outcomes (such as job separations accompanied by demotions) by 2020 by 2.7%. These effects are especially pronounced for older workers. For employees over 40, having modern technical skills reduces the probability of job separations with demotions by 3.9%. These results highlight the interaction between employee age and skill vintage, with important policy implications for workforce reskilling in the face of advancing new technologies.

Mind Over Matter: The Impact of IT Human Capital on Firm Productivity in the Digital Age

Sebastian Steffen
,
Boston College
Wang Jin
,
Stanford University
Erik Brynjolfsson
,
Stanford University

Abstract

In light of the increased prevalence of new information technologies, including cloud computing and artificial intelligence, traditional IT measures based on physical IT capital have become unreliable, while human IT capital has become more important and more easily measurable. New IT technologies have thus, paradoxically, made the measurement of IT capabilities as well as their impact on firm productivity significantly harder than they already were. We create novel IT measurements based on industry-, and firm-level demands for IT skills and occupations from 2010 until 2020. Strong correlations with “official” productivity measures at the industry level validate our approach and suggest their usefulness at the firm level, where no official, reliable measures currently exist. We demonstrate that our measures are robustly associated with higher productivity at both the industry and firm levels, based on a battery of estimation techniques from the productivity literature. Our preferred firm-level estimation implies that a one percent increase in IT skills is associated with a 0.009 percent increase in total sales, which translates to an average gain of $540,000. Our measures are also positively associated with firm innovation, as measured by the total number of patents, citations, and real value of patents, suggesting that IT human capital drives productivity growth through innovation. Our methodology to define these human IT capital measures is general and simple enough to allow for future and backward-compatible extensions. Overall, our results highlight the importance of proper measurement of IT capabilities, and their human capital component in particular, for studying the IT productivity paradox and innovation.

Discussant(s)
Matt Beane
,
University of California-Santa Barbara
Eleanor Dillon
,
Microsoft Research
Sungwoo Cho
,
University of California-Los Angeles
Seth Benzell
,
Chapman University
Felix Koenig
,
Carnegie Mellon University
JEL Classifications
  • O3 - Innovation; Research and Development; Technological Change; Intellectual Property Rights
  • J2 - Demand and Supply of Labor