Digital intelligence transformation measurably improves carbon emission efficiency in China's resource-based cities — a one‑standard‑deviation rise in DIT increases CEE by 0.033 (about 3.96% of the mean). The gains operate through reduced resource misallocation and stronger green innovation, and are largest where green finance and digital infrastructure are more developed.

Main Finding

Digital intelligence transformation (DIT) significantly increases carbon emission efficiency (CEE) in Chinese resource-based cities. A one–standard-deviation increase in the constructed DIT index raises CEE by 0.033, equivalent to a 3.96% increase relative to the sample mean.

Key Points

• Scope: Panel of 110 Chinese resource-based cities over 2013–2022.
• DIT variable: A comprehensive city-level index capturing the depth of digitalization and intelligent-technology adoption (constructed by the authors from available indicators).
• Magnitude: 1 SD rise in DIT → +0.033 CEE (≈ +3.96% of mean).
• Mechanisms: The positive effect operates primarily through
  • reduced resource misallocation, and
  • enhanced green technological innovation.
• Heterogeneity: Effects are stronger in cities with
  • more advanced green finance systems,
  • better digital infrastructure, and
  • cities at mature or regenerative stages of development.
• Robustness: Results reported as robust across the study’s checks (modeling and measurement choices).

Data & Methods

• Data: City-level panel data for 110 resource-based cities in China, 2013–2022.
• DIT index: Constructed by the authors to capture digital + intelligent-technology transformation at the city level (composite index; paper contains construction details).
• Outcome measurement: Carbon emission efficiency (CEE) measured using an SBM-GML approach — a slack-based measure combined with a Global Malmquist–Luenberger framework that accommodates undesirable outputs (CO2 emissions) and allows efficiency comparisons over time.
• Estimation strategy: Two-way fixed-effects panel regression (city and year fixed effects) to estimate the impact of DIT on CEE.
• Mechanism tests: Mediation/auxiliary regressions linking DIT to resource allocation metrics and green-innovation indicators to establish channels.
• Heterogeneity analysis: Subsample regressions by green finance level, digital infrastructure quality, and city development stage.

Implications for AI Economics

• AI/digital-intelligence adoption delivers measurable environmental productivity gains: this study quantifies a ~4% mean improvement in carbon emission efficiency per SD increase in DIT for resource-based cities, showing a concrete macro-regional payoff to intelligent technologies.
• Mechanisms relevant for economic modeling: AI improves allocative efficiency (reducing resource misallocation) and raises returns to green R&D — both channels that can be incorporated into growth and structural-change models to capture AI’s environmental externalities.
• Complementarity matters: The stronger effects where green finance and digital infrastructure are more developed imply non-linear returns and complementarities that AI/economics models should capture (finance, infrastructure, and institution endowments condition AI’s benefits).
• Policy framing: For cities/regions dependent on extractive industries, AI-focused policy should be bundled with green finance and infrastructure investments to maximize carbon-efficiency gains and support regenerative transitions.
• Research directions:
  • External validity: test whether similar magnitudes hold outside Chinese resource-based cities and at firm or sector levels.
  • Causal identification: exploit quasi-experimental variation (instruments, policy shocks) to bolster causal claims about AI-driven efficiency gains.
  • Distributional effects and labor: investigate how DIT-induced reallocation affects employment, wages, and inequality in transition economies.
  • Cost–benefit and dynamic analyses: model the long-run tradeoffs and investment timing for AI and digital infrastructure aimed at decarbonization.

If you’d like, I can (a) extract and summarize the index construction details and robustness checks from the paper, (b) outline a simple structural model that incorporates the channels found here, or (c) suggest empirical strategies to strengthen causal identification for follow-up work.