This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

  1. Home
  2. ASSA Annual Meeting
  3. 2022
  4. Preliminary Program
  5. Using Text Data to Understand the Labor Market
« Back to Results

Using Text Data to Understand the Labor Market

Paper Session

Friday, Jan. 7, 2022 10:00 AM - 12:00 PM (EST)

Hosted By: American Economic Association
  • Chair: Erica Groshen, Cornell University

The Geography of Job Tasks

Enghin Atalay
,
Federal Reserve Bank of Philadelphia
Sebastian Sotelo
,
University of Michigan-Ann Arbor
Daniel Tannenbaum
,
University of Nebraska-Lincoln

Abstract

Working in urban commuting zones (CZs) commands a large earnings premium, and this premium differs significantly by worker skill level. In this paper, we produce new descriptive evidence and introduce new measurement tools to understand the mechanisms behind the urban premium and why it differs by worker skill level. We use the near-universe of job vacancies and develop granular measures of job tasks—based on the natural language employers use, rather than survey-based categories—that allow for differences within occupations and across CZs. We find evidence for three mechanisms behind the earnings premium. First, jobs are more interactive and analytic in urban CZs, even within narrow occupation categories. Second, the computer software requirements of jobs are more intensive in urban CZs. Third, urban workers are more specialized, with less overlap in the sets of tasks performed, within occupations. Furthermore, these differences across CZs are more pronounced for college-educated workers than for non-college workers. We show that job tasks and technologies account for a substantial portion of the urban CZ premium—even within-occupations—and this relationship is stronger for white-collar occupations.

Using Language Models to Understand Wage Premia

Sarah Bana
,
Stanford Digital Economy Lab

Abstract

Does the text content of a job posting predict the salary offered for the role? There is ample evidence that even within an occupation, a job's skills and tasks affect the job's salary. Capturing this fine-grained information from postings can provide real-time insights on prices of various job characteristics. Using a new dataset from Greenwich.HR with salary information linked to posting data from Burning Glass Technologies, I apply natural language processing (NLP) techniques to build a model that predicts salaries from job posting text. This follows the rich tradition in the economics literature of estimating wage premia for various job characteristics by applying hedonic regression. My model explains 87 percent of the variation in salaries, 26 percent (18 percentage points) over a model with occupation by location fixed effects. I apply this model to the question of online certifications by creating counterfactual postings and estimating the salary differential. I find that there is substantial variation in the predicted value of various certifications. As firms and workers make strategic decisions about their human capital, this information is a crucial input.

Eight Decades of Changes in Occupational Tasks, Computerization and the Gender Pay Gap

Andre Assumpcao
,
Harvard University
Dario Diodato
,
Joint Research Centre of the European Commission
Ljubica Nedelkoska
,
Harvard University
Shreyas Gadgin Matha
,
Harvard University
James McNerney
,
Harvard University
Frank Neffke
,
Harvard University

Abstract

We build a new longitudinal dataset of job tasks and technologies by transforming the U.S. Dictionary of Occupational Titles (DOT, 1939 - 1991) and four books documenting occupational use of tools and technologies in the 1940s, into a database akin to, and comparable with its digital successor, the O*NET (1998 - today). After creating a single occupational classification stretching between 1939 and 2019, we connect all DOT waves and the decennial O*NET databases into a single dataset, and we connect these with the U.S. Decennial Census data at the level of 585 occupational groups. We use the new dataset to study how technology changed the gender pay gap in the United States since the 1940s. We find that computerization had two counteracting effects on the pay gap - it simultaneously reduced it by attracting more women into better-paying occupations, and increased it through higher returns to computer use among men. The first effect closed the pay gap by 3.3 pp, but the second increased it by 5.8 pp, leading to a net widening of the pay gap.

Augmented Intelligence

Erik Brynjolfsson
,
Stanford Digital Economy Lab
Lindsey Raymond
,
Massachusetts Institute of Technology

Abstract

How does technology affect the organization and dissemination of knowledge in production? Production knowledge is often tacit and embodied in individual workers and managers, and therefore technology that changes the ability to encode and disseminate knowledge, such as artificial intelligence, might affect productivity and organization of production.

We study a combination of experimental and staggered deployment of an AI conversational intelligence chatbot across 3,00 agents and 4 million tech support conversations in a Fortune 100 software company that gives technical support agents real-time suggestions on what to say. First, we find that the AI increases agent productivity on both efficiency (average time per call) and quality (call resolution and customer satisfaction), with the lowest skilled workers seeing the largest relative gains. Within teams, the gap between top and bottom performing agents falls after adoption, with ex-ante less skilled agents seeing the largest gains, particularly in problem solving ability. Consistent with the AI increasing knowledge diffusion, we see dissemination of “best practices” after AI deployment, particularly across firm, geographic and team boundaries in the text of technical support conversations. As a result of the reduced cost of knowledge acquisition for the technical support agents, at the team level, the average number of agents assigned to one manager increases and the average number of distinct topics solved by each agent also rises. Together, these results have implications for the role of AI in affecting the organization of production and boundary of the firm.

Discussant(s)
Adrien Bilal
,
Harvard University
Stephen Hansen
,
Imperial College London
Anna Salomons
,
Utrecht University
Avi Goldfarb
,
University of Toronto
JEL Classifications
  • J2 - Demand and Supply of Labor
  • J6 - Mobility, Unemployment, Vacancies, and Immigrant Workers