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

'Generate' the Future of Work through AI: Empirical Evidence from Online Labor Markets

Liu, Xu, Li, Tan

2023arXiv17 citations

Observational labor marketInformation SystemsCausal

LLM / Generative AIPlatforms / gig economySoftware / codingWriting / contentAugmentation vs. substitutionTraining / upskillingOccupational mobility

View Repository DOI: 10.2139/ssrn.4529739

Summary

Liu et al. use difference-in-differences analysis on comprehensive data from a leading online labor platform (September 2021 - August 2023) to study how ChatGPT's launch affected demand, supply, competition, and skill transitions across submarkets classified by LM-AIOE exposure.

Main Finding

ChatGPT caused a 22% decline in job postings and 16.8% decline in bids in exposed submarkets, intensifying competition; freelancers transitioned toward programming jobs, with high-skilled freelancers driving most of the skill transition and realizing 11.75% higher earning efficiency.

Primary Datasets

Leading online labor platform data (all job postings, bids, transactions, freelancer data)

Secondary Datasets

O*NET

Other

Google Trends

Other

LM-AIOE (Language Model AI Occupational Exposure Index); O*NET occupation descriptions; Google Trends data

Key Methods: Difference-in-differences (DiD) and triple-difference (DDD) specifications using natural language processing to cluster jobs into submarkets and LM-AIOE to define treatment; interrupted time series analysis and propensity score matching for robustness.
Sample Period: 2021-2023
Geographic Coverage: Global online labor platform (92% of freelancers from developing economies)
Sample Size: 1.6 million jobs, 2 million freelancers, 132,260 active freelancers (pre- and post-treatment), 1,083 submarkets
Level of Analysis: Individual, Occupation, Task
Occupation Classification: LM-AIOE occupation mapping; platform skill tags clustered into submarkets
Industry Classification: None

Notes

arXiv:2308.05201 [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications. [Claude classification]: This paper examines the natural experiment created by ChatGPT's launch on November 30, 2022. Treatment assignment uses LM-AIOE scores mapped to platform skill tags via semantic similarity, then aggregated to submarkets using HDBSCAN clustering. The paper documents both displacement effects (demand-side) and skill transition effects (supply-side), with heterogeneity analysis showing high-skilled freelancers drive the transition to programming. Main robustness checks include placebo tests, propensity score matching, interrupted time series, and alternative specifications.