Cities that better recombine digital and green knowledge achieve higher total-factor carbon efficiency; institutional strength, innovation capacity and a larger tertiary sector amplify these carbon-efficiency gains.

Main Finding

Higher levels of Digital-Green Convergent Innovation (DGCI) — conceptualized as a quality‑weighted process of recombining digital and green knowledge — significantly improve regional Total Factor Carbon Efficiency (TFCE) in Chinese cities. These carbon‑efficiency gains are amplified by institutional conditions (e.g., digital infrastructure policy and environmental fiscal spending) and are larger in regions with stronger innovation capacity and more developed tertiary sectors.

Key Points

• Conceptualization: DGCI is framed not as mere co-occurrence of “digital” and “green” technologies but as a knowledge‑factor recombination process whose impact depends on the quality and scarcity (non‑ubiquity) of recombined elements.
• Measurement: The authors construct a novel DGCI index that jointly captures the diversity and non‑ubiquity of digital–green technological recombination (aiming to distinguish high‑value convergence from low‑impact overlap).
• Empirical strategy: To address simultaneity, reverse causality, and selection bias common in two‑step approaches, the paper uses an endogenous stochastic frontier (one‑step) model to estimate TFCE and its determinants jointly.
• Main empirical results: Using a panel of 227 Chinese cities (2010–2021), higher DGCI is associated with significant improvements in TFCE. Institutional moderators strengthen this effect. Heterogeneity tests show larger DGCI→TFCE effects where regional innovation capacity and tertiary‑sector development are higher.
• Mechanisms (theoretical): DGCI improves TFCE via richer digital feedback loops, real‑time optimization, co‑specialization of knowledge, and improved coordination of production and governance — i.e., recombined digital knowledge embedded into green R&D reshapes innovation trajectories toward higher carbon efficiency.
• Methodological critique: The paper highlights limitations of patent‑overlap and simple diversity measures and argues for quality‑weighted measures of convergence plus one‑step frontier estimation to avoid biased inference.

Data & Methods

• Data: Panel dataset covering 227 Chinese cities over 2010–2021. (DGCI constructed from technology/innovation indicators capturing digital‑green recombination; TFCE measured via a production‑frontier framework that accounts for multiple inputs, desirable outputs, and carbon emissions.)
• DGCI index: A newly proposed index that incorporates both diversity of recombined elements and their non‑ubiquity (quality/scarcity) to better capture meaningful digital‑green convergence.
• Econometric method: Endogenous Stochastic Frontier Analysis (ESFA) — a one‑step approach that jointly estimates the TFCE frontier and the effects of DGCI and other determinants, addressing endogeneity, simultaneity, and selection concerns that plague two‑step DEA/SFA regressions.
• Robustness and heterogeneity: The study runs moderator analyses (institutional variables such as digital infrastructure policy and environmental fiscal spending) and subgroup tests by regional innovation capacity and tertiary sector development.

Implications for AI Economics

• Measurement of AI‑Green Convergence: For research that studies AI’s environmental impacts, measuring convergence should go beyond co‑occurrence and capture the quality/scarcity of AI knowledge embedded in green technologies (a quality‑weighted index like DGCI can better predict meaningful outcomes).
• Endogenous diffusion of AI: Regions that already have higher TFCE or innovation capacity may both attract and generate higher‑quality AI→green recombination; empirical work must account for this endogeneity (ESFA or other joint estimation methods are recommended).
• Policy levers for AI‑driven decarbonization: Policies that build digital infrastructure, target environmental fiscal spending, and strengthen regional innovation systems can magnify the carbon‑efficiency gains from AI/digital adoption in green sectors. Blanket AI diffusion is unlikely to yield uniform benefits without complementary institutions and sectoral capacity.
• Heterogeneous returns to AI integration: The climate‑efficiency payoff of embedding AI into green technologies will vary by region — larger where tertiary sectors and local innovation ecosystems can translate AI capabilities into high‑quality recombinations. This argues for tailored regional strategies rather than one‑size‑fits‑all AI deployment.
• Research design guidance: AI economics analyses of environmental impacts should (a) explicitly model knowledge recombination (not just adoption), (b) weight for quality/rarity of recombined knowledge, and (c) use estimation strategies robust to reverse causality and selection (e.g., one‑step frontier models, instrumental approaches).
• Mechanism attention: Empirical and theoretical work should unpack the operational channels (real‑time optimization, digital twins, predictive control, inter‑firm spillovers) by which AI‑embedded recombination reduces emissions intensity for a given output level.

If you want, I can (a) extract the DGCI construction algorithm from the manuscript (patent/technology variables and weighting), (b) sketch an empirical replication plan using open data, or (c) produce a short list of policy recommendations for cities seeking to maximize AI‑enabled low‑carbon gains.