China’s digital-economy growth appears to absorb more low- and mid-skilled workers at the provincial level while also drawing in high-skilled workers, but estimated gains are larger for lower-educated groups — suggesting regional agglomeration remains a key driver of job absorption for less-skilled labor.

Main Finding

The authors construct a multidimensional digital-economy index for 31 Chinese provinces (2013–2022) using a combined global principal component analysis (GPCA) + global entropy weighting approach and estimate spatial panel models to link regional digital-economy development to the composition of labor by education level. They find that (1) digital-economy development is geographically concentrated (east > central > west) and heavily driven by digital infrastructure; (2) digitalization has heterogeneous effects on employment: it significantly expands employment for low- to medium-educated workers (via economic expansion and agglomeration effects) and also increases the concentration of highly educated workers but to a smaller extent; and (3) there are notable spatial spillovers — a region’s digital-economy development affects neighboring regions’ labor-structure outcomes. The authors recommend coordinated regional infrastructure, graded industrial transfers, focused upskilling/vocational training, and policies to attract high-skilled talent.

Key Points

• Index construction and dominant drivers
  • Digital-economy index built from 10 indicators across three dimensions: digital inputs, digital outputs, and digital infrastructure.
  • Digital infrastructure accounts for the largest share of the index (≈53%), followed by inputs (≈28%) and outputs (≈19%).
• Measurement approach
  • Combined objective weighting: arithmetic average of GPCA-derived weights and global entropy-method weights to improve robustness.
  • KMO for PCA = 0.78; three principal components retained, cumulative variance explained ≈82.8% (PC1 54.3%, PC2 18.6%, PC3 9.9%).
• Empirical strategy
  • Sample: panel of 31 Chinese provincial-level regions, years 2013–2022.
  • Models compared: spatial Durbin model (SDM), spatial Durbin error model (SDEM), spatial lag model (SLM), and spatial error model (SEM); selection by AIC.
  • Final reported spatial specifications used for inference: spatial Durbin and spatial lag models (accounting for direct and spillover effects).
• Main empirical results
  • Higher regional digital-economy levels are positively associated with employment scale for low- and medium-educated workers — interpreted as agglomeration-driven absorption and expanded employment opportunities.
  • Digital-economy development, GDP growth, and population agglomeration also increase concentration of highly educated workers, but the coefficient magnitude is smaller than for lower-education groups.
  • Significant spatial spillovers imply that digital development in one province influences neighboring provinces’ employment structures.
• Policy recommendations highlighted
  • Strengthen inter-regional coordination of digital infrastructure and promote graded (tiered) industrial transfers.
  • Implement targeted vocational/upskilling programs for low- to medium-skilled workers.
  • Attract and retain high-skilled talent via industry–academia–research linkages and urban-ecosystem improvements.
  • Use spatial governance tools to mitigate negative spillovers and attend to education, trade, and urbanization policies.

Data & Methods

• Data
  • Geographic scope: 31 provincial-level regions in China.
  • Time span: 2013–2022.
  • Primary data sources: China National and provincial statistical yearbooks, CSMAR database, and other official digital-economy statistics.
• Digital-economy indicator set (representative)
  • Digital inputs: proportion of information-industry personnel; R&D expenditure of large enterprises; digital inclusive finance index.
  • Digital outputs: number of AI enterprises; number of patent applications; per-capita telecommunications revenue.
  • Digital infrastructure: internet broadband access rate; mobile-phone penetration; mobile users per 100 people.
• Index construction
  • Standardization, then two objective weighting procedures: GPCA and global entropy method; final weights = arithmetic mean of the two sets.
  • PCA diagnostics: KMO = 0.78; three components retained (cumulative variance ≈82.8%).
  • Result: infrastructure dimension dominates the composite index.
• Econometric strategy
  • Baseline panel regressions of labor-employment structure by education level on the digital-economy index, with standard control variables (e.g., GDP/economic growth, population agglomeration/density, urbanization, educational investment, trade openness — as reported/used in robustness checks).
  • Spatial econometric analysis: compared SDM, SDEM, SLM, and SEM using AIC; proceeded with spatial Durbin and spatial lag specifications to capture direct and neighboring-region (spillover) effects.
  • Interpretation emphasizes both direct effects (within-region) and indirect/spillover effects (between neighboring regions).

Implications for AI Economics

• Measurement implications
  • Composite indices that combine GPCA and entropy weighting can robustly capture multi-dimensional digital/AI ecosystem development; infrastructure often dominates regional digitalization measures and thus should be carefully tracked in AI-economy studies.
  • Spatial heterogeneity matters: evaluations of AI-related policies or technology diffusion should explicitly model spatial dependence and spillovers.
• Labor-market theory and policy
  • Evidence that digitalization (including AI-related activity) can expand employment opportunities for low- and medium-educated workers via agglomeration and industry expansion challenges simplistic narratives of universal job-displacement; instead, effects are heterogeneous by region and skill.
  • Simultaneous, but smaller, concentration of high-skilled workers implies continuing skill-biased localization of high-end AI jobs — policies must balance attraction of top AI talent with measures to broaden skill-upgrading opportunities across the workforce.
• Research agenda for AI economics
  • Move beyond regional aggregate indices to micro-level measures of AI penetration and firm digital maturity (e.g., adoption of AI systems, automation intensity, platform participation) to unpack mechanisms (substitution vs. complementarity).
  • Use spatial panel and multi-level causal methods to identify how AI diffusion across space alters local labor compositions, wages, and sectoral reallocations.
  • Investigate policy interventions (vocational training, regional infrastructure investment, targeted subsidies) in quasi-experimental frameworks to assess how they mediate AI’s labor-market impacts.
• Policy design considerations
  • Spatial coordination of digital/AI infrastructure reduces unequal diffusion and negative spillovers; graded industrial transfer (moving labor-intensive operations to lower-cost regions) can enlarge employment for lower-educated workers but requires complementary training and social protections.
  • To foster inclusive AI-driven growth, combine investments in digital infrastructure with large-scale vocational training, incentives for industry–academia collaboration, and urban policies that improve absorptive capacity for high-skilled AI talent.