This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.

The Diffusion of Disruptive Technologies

Bloom, Hassan, Kalyani, Lerner, Tahoun

2025Quarterly Journal of Economics5 citations

Observational labor market

Automation / RobotsMachine Learning (pre-LLM)AI (General)Platforms / gig economyGeneral automation

View Repository DOI: 10.2139/ssrn.3880232

Summary

Bloom, Hassan, Kalyani, Lerner, and Tahoun use textual analysis of patents, job postings, and earnings calls to identify 1,286 new technologies developed since 1976 and trace their geographic, occupational, and skill diffusion patterns across U.S. labor markets from 2010-2019.

Main Finding

New technologies originate in highly concentrated geographic locations (56% of the most impactful technologies emerge from Silicon Valley and the Northeast Corridor), spread slowly across regions (taking over 50 years to fully disperse), initially require high skills (57% require college degrees) that decline gradually over time, and pioneer locations retain persistent advantages in high-skill jobs for decades.

Primary Datasets

USPTO Patent

Other

Lightcast (formerly Burning Glass)

Labor outcomes

USPTO patents; Burning Glass job postings

Secondary Datasets

Earnings call transcripts; SEC filings

Key Methods: Natural language processing and textual analysis of patents, Wikipedia pages, earnings call transcripts, and online job postings; construction of technology emergence measures and pioneer location identifiers; descriptive panel regressions tracking diffusion patterns over time and space
Sample Period: 2000-2020
Geographic Coverage: US
Sample Size: Approximately 3 million patents, 200 million job postings, 321,373 earnings calls, 1,899 technology bigrams (1,286 unique technologies)
Level of Analysis: Occupation, Industry, Region, Firm
Occupation Classification: SOC (6-digit)
Industry Classification: NAICS (4-digit)
Replication Package: Yes

Notes

Published in QJE (2025); NBER WP 28999 [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions. [Claude classification]: Published in QJE 2025 (originally NBER WP 28999, 2021). Paper identifies new technologies through intersection of patent text, Wikipedia, earnings calls, and job postings. Key methodological innovation is the systematic identification of technology phrases using NLP. Main analysis focuses on 276 'economically impactful' technologies (mentioned in 100+ earnings calls). Finds extreme geographic concentration of innovation origins and persistent regional advantages lasting decades. Human audit validates 73% accuracy of Wikipedia technology filter. Multiple robustness checks using unigrams, trigrams, alternative emergence year definitions.