Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 31 of the 32 references it cites were found in a registry or an index, and 1 are not checked yet; none of the 63 sources its claims cite is checked for support yet.

The Expanding Elite Premium: Task‐Biased Technological Change and Higher Education Stratification in China

Open access
Sun, Chen, Yang, Juan · 2026 · The Manchester School

Abstract

This paper examines how wage premia associated with institutional tiers in Chinese higher education evolved between 2013 and 2023, a period of continued higher education expansion that also encompasses the early public diffusion of generative AI. Drawing on three waves of the China Household Income Project, 5.19 million online job postings, and an occupation‐level generative AI exposure index, we document a substantial widening of the wage premium associated with elite universities within higher education. Relative to junior college graduates, the log hourly wage premium for elite university graduates rose from 0.17 in 2013 to 0.35 in 2023, corresponding to an hourly wage advantage of approximately 42% in 2023, roughly twice the corresponding premium for ordinary university graduates, which remained broadly stable over the same period. RIF‐based Oaxaca–Blinder decompositions show that an increasing share of the ordinary–elite wage gap cannot be accounted for by observable worker composition and is instead associated with differences in estimated wage structures, with the widening concentrated at the upper end of the wage distribution. Between 2018 and 2023, occupations with greater AI exposure experienced stronger relative growth in recruitment and posted wages for non‐routine than for routine work. Matching the exposure measure to the CHIP data further shows that the association between AI exposure and non‐routine occupational allocation is more pronounced for elite graduates, while the elite wage differential within non‐routine occupations also widened in 2023. Taken together, the evidence is consistent with persistent stratification in educational supply and task‐biased changes in labour demand jointly contributing to the widening elite‐university premium.

Checked 2026-11-17 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

Using CHIP 2013/2018/2023, 5.19 million online job postings, and an occupation-level generative AI exposure index, the paper documents a widening elite-university wage premium in China: the elite log hourly wage premium over junior college rose from 0.17 to 0.35, about 42% in 2023, while the ordinary-university premium stayed broadly stable. The widening is concentrated at the upper tail and associated with wage-structure differences, and the demand-side evidence links greater AI exposure to stronger non-routine recruitment/wage growth and to elite graduates’ greater non-routine occupational allocation and conditional wage advantage.

Key findings

  • Relative to junior college graduates, the elite-university log hourly wage premium rose from 0.17 in 2013 to 0.35 in 2023, corresponding to an approximately 42% hourly wage advantage in 2023, roughly twice the ordinary-university premium. The ordinary-university premium remained broadly stable: 0.160, 0.124, and 0.184 in 2013, 2018, and 2023, or about 17.3%, 13.2%, and 20.2%.
  • RIF-based Oaxaca-Blinder decompositions show the ordinary-elite mean wage gap increased from 6.532 in 2013 to 17.997 in 2023; the coefficient effect rose from 4.155 to 14.262, increasing its share from 63.6% to 79.2%. The P90-P10 differential between ordinary- and elite-university graduates rose from 7.218 to 46.707, of which 43.107, or more than 92%, was accounted for by the coefficient effect in 2023. Quantile estimates show the elite premium at the 50th and 90th percentiles rising from 0.202 and 0.301 in 2013 to 0.354 and 0.862 in 2023.
  • Between 2018 and 2023, occupations with higher AI exposure experienced stronger relative growth in recruitment and posted wages for non-routine than routine work. The PPML interaction between non-routine status and standardised AI exposure is 1.614 (significant at 5%); the posted-wage interaction is 204.14 yuan (significant at 1%), implying a one-standard-deviation increase in AI exposure raises posted monthly wages of non-routine occupations by about 204 yuan relative to routine occupations. Annual estimates show the non-routine–routine AI-exposure slope difference at 0.001 in 2022, 0.034 in 2023, and 0.115 (p = 0.012) in 2024.
  • Matching the exposure measure to CHIP shows the elite × AI exposure × 2023 coefficient for non-routine occupational allocation is 0.341 (significant at 10%), while the elite × non-routine × 2023 wage coefficient is 0.367 (significant at 5%), implying the conditional elite wage differential in non-routine work increased by approximately 44.3% between 2018 and 2023 relative to junior-college graduates.

Methods

  • Design combines repeated cross-sectional individual wage analysis with demand-side job-posting analysis. CHIP waves cover 2013, 2018, and 2023; the sample is urban residents aged 18–65 with junior college or undergraduate degrees and valid earnings/institution information, yielding 1328, 3916, and 889 observations after cleaning. Job-posting data cover the full 51job platform from January 2014 to December 2024, retaining full-time positions requiring junior college or above and about 5.19 million valid postings after cleaning. The AI exposure measure is the China LLM Exposure Index for 2018–2024, covering roughly one hundred occupations and standardised within year.
  • Main analyses include extended Mincer log hourly wage equations with institutional-tier dummies and province/industry fixed effects; two-stage least squares using provincial college admission rate and number of higher-education institutions as instruments; quantile regressions; and RIF-based Oaxaca-Blinder decompositions. Demand-side models use PPML for vacancy counts and OLS for posted monthly wages with non-routine × AI exposure × post-2023 interactions; individual-level triple-interaction models examine non-routine occupational allocation and log wages.

Limitations

  • The paper states that temporal coverage limits the analysis to comparable observations around 2018 and 2023, so not all contemporaneous influences can be separately identified; results are interpreted as evidence of task-differentiated labour-market adjustment associated with AI exposure, not as a causal estimate of the standalone effect of generative AI.
  • The paper states that CHIP 2023 records institution names rather than predefined tiers, reducing effective sample size and potentially introducing selection; inverse-probability weighting and entropy balancing preserve the widening elite premium after balancing observed reporting differences, but unobserved determinants of reporting cannot be adjusted. The AI exposure index measures potential exposure rather than realised adoption, and the task taxonomy abstracts from within-occupation heterogeneity. Comparable field-of-study information is unavailable across CHIP waves, so tier differentials include field composition.
  • The paper also notes that the coefficient effect should not be interpreted as a pure return to elite education because it may capture unobserved ability, field of study, firm and occupational sorting, and matching quality. The text makes evident that online job postings over-represent urban, highly educated, middle-to-high-income occupations, though the paper argues this coverage aligns with the study’s focus and that postings are used for relative changes rather than labour-market aggregates.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1111/manc.70080

Cites

31 of 32 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 0 Not valid · 47 Not checked · 201 Own results and interpretations

103 more claims

Create a free account to see all 248 claims and to verify the 36 references still open.

See all claims — free account
  • It may state two separate claims as one. 99
  • Probably not a substantial claim. 66

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 31 references

Research works printed but not linked: 19

  1. Acemoglu D., 2011, Handbook of Labor Economics, 1043
  2. Cai F., 2011, Wage Growth, Wage Convergence and the Lewis Turning Point, Economic Perspectives, 9
  3. Ding X., 2012, Returns to Education at Different Levels for Urban Residents in China, 2002–2009, Peking University Education Review, 10, 73
  4. Fang T., 2026, AI and Labor Market Outcomes: Evidence from China. IZA Discussion Paper No. 18740
  5. Goldin C., 2008, The Race Between Education and Technology
  6. He Q., 2026, Generative Artificial Intelligence and the Structural Differentiation of Firm Labour Demand: Evidence From a Quasi‐Natural Experiment Based on the Release of ChatGPT, Journal of Shanghai University of Finance and Economics, 28, 45
  7. He Y., 2009, Empirical Analysis of the Changing Returns to Education Under Educational Expansion, Chinese Journal of Population Science, 44
  8. Humlum A., 2025, Large Language Models, Small Labor Market Effects, 33777
  9. Li L., 2021, The Employment Effects of Robots: Mechanism and Evidence From China, Management World, 37, 104
  10. Li S., 2003, Long‐Term Changes in Returns to Education in Urban China, Social Sciences in China, 58
  11. Li Y., 2024, How can Education Change One’s Destiny? Evidence on the Long‐Term Returns to Elite Education, China Economic Quarterly, 24, 1394
  12. Luo C., 2018, Long‐Run Changes in the Returns to Education in Urban China, Social Sciences in Chinese Higher Education Institutions, 159, 97
  13. Peng S., 2017, The Inverted U‐shaped Evolution of the College Wage Premium in China: Evidence From CHIP 1988–2013, South China Journal of Economics, 37
  14. Wang Y., 2020, How the Rise of Robots has Affected China’s Labour Market: Evidence From Manufacturing Listed Firms, Economic Research Journal, 55, 159
  15. Webb M., 2020, The Impact of Artificial Intelligence on the Labor Market
  16. Xu S., 2010, Technical Progress, Returns to Education and Income Inequality, Economic Research Journal, 45, 79
  17. Yu L., 2021, Industrial Robots, Work Tasks and Non‐routine Skill Premium: Evidence From a Matched Employer–Employee Survey in Chinese Manufacturing, Management World, 37, 47
  18. Zhang D., 2025, The Measurement of AI Exposure and Its Impact on Labour Demand in China: Evidence From Large Language Models, Management World, 41, 59
  19. Zhou Y., 2020, From College to Elite College: Heterogeneous Returns to Higher Education and Social Sorting Under Expansion, Educational Research, 41, 86

Cited by

Still being checked.

Metadata

Authors
Sun, C., & Yang, J.
Affiliations
Beijing Normal University (CN)
Corresponding author
Yang, Juan
Venue
The Manchester School
Publisher
Wiley
Published
2026-11-13
ISSN
  • 0025-2034 (ISSN-L)
  • 1463-6786
  • 1467-9957
OpenAlex
W7231816176
Cited by
0
References
31
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
Manchester School (publishedVersion)
Topics
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
  • Social Sciences › Sociology and Political Science › Intergenerational and Educational Inequality Studies
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Education and Labor Relations
Keywords
higher education stratification, wage premium, generative AI, educational supply, labor demand, China Household Income Project, online job postings

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-17, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-17 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: The Expanding Elite Premium: Task‐Biased Technological Change and Higher Education Stratification in China. https://app.cochat.ai/papers/10.1111/manc.70080
The paper
Sun, C., & Yang, J. (2026). The Expanding Elite Premium: Task‐Biased Technological Change and Higher Education Stratification in China. The Manchester School. https://doi.org/10.1111/manc.70080

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.