Spatial Effects and Mechanisms of the Digital Economy and Industrial Structure on Urban Carbon Emissions: Evidence from 274 Chinese Cities

Abstract

As China advances toward its “Dual Carbon” goals, clarifying the role of the digital economy (DE) in reducing urban carbon emissions is of growing importance. This study uses panel data from 274 Chinese prefecture-level cities (2011–2022) and applies benchmark regression, the Spatial Durbin Model (SDM), two-regime SDM, threshold analysis, and mediation effect modeling to examine the impact of the DE on carbon emission intensity (CEI) and its spatial spillover effects. Results show that the DE significantly reduces CEI through both direct and indirect channels. Spatial analysis reveals that the DE’s spillover effect is most pronounced within a 500 km range. Regionally, the DE has a stronger inhibitory effect on CEI in eastern and western regions, while its effect in the central region is weaker or even reversed, likely due to reliance on carbon-intensive industries. Resource-based cities exhibit stronger spatial spillovers than non-resource-based ones, suggesting greater potential for DE-driven low-carbon transitions. A threshold effect is also identified at a DE index value of 0.0326, beyond which the marginal benefits decline. Pathway analysis indicates that while the DE improves production efficiency, it does not significantly promote green, high-value-added transformation, partially masking its carbon reduction effects. These findings highlight the need for tailored regional strategies to enhance the low-carbon potential of the DE.

1. Introduction

Amid global climate initiatives and China’s pursuit of its “Dual Carbon” targets—peaking carbon emissions by 2030 and achieving carbon neutrality before 2060—urban areas, as major centers of economic activity and carbon emissions, face the critical challenge of reconciling economic growth with emission reduction (Xi et al., 2024; Huang et al., 2022; Zi et al., 2023). With the rapid growth of China’s urban economy, which has elevated the country to the world’s second-largest economy, energy consumption has surged, leading to frequent environmental problems and a sharp rise in greenhouse gas emissions, making China one of the world’s the largest carbon emitters (X. Zheng et al., 2020; An et al., 2021). China’s economy is undergoing a critical transition toward high-quality and sustainable growth, and achieving green, low-carbon development has thus become a strategic imperative in urban sustainable planning (Y. Zhang, 2022).The DE, as a catalyst for technological innovation and industrial transformation, offers new pathways for urban low-carbon transition. However, the rapid expansion of data centers, digital infrastructure, and energy-intensive platforms may contribute to increased carbon footprints (Hertwich, 2021). These conflicting dynamics underscore the need to investigate whether the DE acts as a driver of carbon reduction or an inadvertent source of emissions. Consequently, investigating the influence of the DE on CEI in Chinese cities holds both theoretical relevance and practical value. From an applied perspective, clarifying how the DE can drive low-carbon transformation, along with its spatial spillover effects, provides a solid foundation for developing targeted, effective emission-reduction policies and regional strategies.

Existing studies have extensively examined the role of the DE in reducing carbon emissions (e.g., W. Zhang et al., 2022; X. C. Zhao et al., 2023; Liu et al., 2023; Jing et al., 2023; Du & Wang, 2024; C. Li & Zhou, 2024), with a primary focus on efficiency improvements, industrial upgrading, and green innovation. Some scholars have applied spatial econometric models to investigate spatial heterogeneity (J. Zheng et al., 2023; Yuan et al., 2025; Ding et al., 2025), while others have explored moderating effects such as green finance (Jin et al., 2025). In addition, a number of studies have approached the issue from an international perspective, analyzing the relationship between the DE and carbon emissions across countries (Dong et al., 2022; Z. Zhang et al., 2024; X. F. Zhang et al., 2025). However, few have integrated spatial econometric approaches with threshold modeling and multi-regional heterogeneity analysis. Moreover, the mechanisms through which the digital economy generates spatial spillovers and decay effects across geographic distance remain insufficiently examined. This study addresses these gaps by adopting a comprehensive framework that incorporates the two-regime SDM, spatial mediation modeling, and threshold regression.

A summary review of the existing literature highlights the significant progress that has been made in terms of the DE influencing carbon emissions, laying a solid theoretical basis for this study. However, several issues remain. First, most prior work relies on single econometric models, limiting the ability to capture complex spatial dynamics and heterogeneity across regions. Second, relatively little attention has been paid to the spatial spillover and decay effects of the digital economy, leaving unanswered questions about how far such benefits extend geographically. Third, while international evidence has emphasized the dual role of digitalization in reducing emissions and promoting industrial transformation, less is known about how these dynamics unfold in developing economies such as China, where structural inertia and uneven regional development pose unique challenges.

Therefore, the possible marginal contributions are as follows. First, a multi-dimensional spatial analysis framework is introduced that incorporates both the SDM and a two-regime SDM, enabling us to capture heterogeneity between resource-based and non-resource-based cities. Second, we are the first to explore the spatial decay boundary of digital economy spillovers across distance bands, finding that effects diminish beyond 900 km. Third, by identifying a threshold effect of the digital economy index (at 0.0326), we reveal that benefits plateau or decline once digital development reaches a certain level. Lastly, this study performs a comprehensive mediation and masking effect analysis to disentangle the roles of industrial structure upgrading, finding that digital development may inhibit green transformation in certain regions, a nuance not previously explored.

The remainder of this paper is structured as follows. Section 2 constructs the theoretical framework and presents the research hypotheses concerning the DE’s influence on carbon emission intensity, including spatial spillovers and mediating mechanisms. Section 3 outlines the research design, introduces the variables and data sources, and explains the empirical models employed, such as the benchmark regression, SDM, two-regime SDM, the mediation effect model, and threshold regression. Section 4 presents and discusses the empirical results, including robustness checks, spatial heterogeneity analysis, and pathway decomposition. Section 5 provides a detailed discussion of the findings and their implications. Finally, Section 6 concludes the study and offers targeted policy recommendations to support urban low-carbon transformation in the context of digital development.

2. Theoretical Analysis and Research Hypotheses

2.1. The Mechanisms Through Which the DE Affects the CEI

The DE primarily influences the CEI through the following three mechanisms.

First, it does so by enhancing production efficiency. The DE promotes the widespread use of technologies such as the Internet of Things, cloud computing, and cloud data, facilitating the automation and intelligent control of production processes. By utilizing smart production systems, companies can better manage energy consumption, reduce resource waste, and ultimately contribute to carbon emission reduction (Z. Li & Wang, 2022).

Second, it influences the CEI by optimizing resource allocation. The DE enhances resource allocation efficiency across sectors such as energy, logistics, and raw materials by leveraging internet connectivity and real-time information flow. Through digital platforms, supply-chain management efficiency is greatly improved, reducing unnecessary energy use and carbon emissions during transportation. Additionally, digital technologies allow businesses and governments to more accurately forecast demand and allocate resources, helping to prevent overproduction and excessive resource consumption (Thompson et al., 2013). Third, it influences the CEI by facilitating the development and deployment of green technologies. The continuous evolution of digital technology has provided strong backing for the development and application of green technologies and promotes the innovation of related technologies. It has optimized the process of technology development, enhanced the efficiency of green technology implementation, and expedited the widespread adoption of these technologies. Through digital technology, the application of green technologies has become more precise, accelerating their adoption across industries and fostering sustainable industrial transformation.

Therefore, this study proposes Hypothesis (1): the DE reduces the CEI.

2.2. The Spatial Spillover Effects of the DE on Regional CEI

The effect of the DE on regional CEI can generate spatial spillover effects due to the following pathways.

The first is interregional cooperation. The DE accelerates economic integration and cross-regional collaboration, enabling carbon reduction achievements in one region to spill over into neighboring areas through the flow of technology, industries, and capital, thus generating positive emissions reduction effects. Second is technology and knowledge spillover (Zhou et al., 2024). The DE accelerates technological innovation and the dissemination of green technologies, facilitating the widespread adoption of low-carbon and green technologies. These innovations not only impact the local area but also extend to neighboring regions through interregional spillovers, boosting local green productivity, improving energy efficiency, and reducing carbon emissions (Chen & Jiang, 2023; Y. S. Li et al., 2022; Shao et al., 2022). Third is industrial chain complementarity. Digital transformation helps upgrade traditional high-carbon industries to low-carbon and high-efficiency models. This shift can affect neighboring regional industries by influencing the industrial chain or fostering cross-regional cooperation, encouraging the adoption of carbon reduction measures.

Based on these mechanisms, this study proposes Hypothesis (2): the DE has spatial spillover effects on CEI.

2.3. The Mediating Effect of Industrial Structure Upgrading Between the DE and CEI

The industrial structure upgrade serves as a mediator between the DE and CEI, primarily in two ways.

The first is by driving the shift to higher-end industries. The DE accelerates the dissemination of information and fosters technological innovation, leading industries to evolve toward sectors with higher value-added and more advanced technologies (Yan et al., 2022). These advanced industries demonstrate higher production efficiency and lower CEI. As the DE develops, traditional industries are transitioning to sectors like technology, services, and green manufacturing, directly reducing carbon emission intensity. Additionally, high-tech industries, equipped with advanced machinery and green technologies, use energy more efficiently, further decreasing carbon emissions. This structural transformation supports the expansion of low-carbon industries, thereby contributing to emission reduction.

Second, the industrial structure upgrade serves as a mediator between the DE and CEI by encouraging the adoption of green technologies and clean energy. As industries shift towards green and low-carbon models, the DE accelerates the implementation of green technologies and clean energy. By utilizing digital technologies, energy consumption is optimized, reducing reliance on traditional fossil fuels. The DE encourages the widespread use of clean energy, decreasing dependence on high-carbon sources like coal and oil, thereby resulting in a carbon emissions reduction. Furthermore, the DE enables precise energy management and smart scheduling, enhancing energy efficiency and further lowering carbon emissions (Zhou et al., 2024; Gu et al., 2023).

In summary, this study proposes Hypothesis (3): industrial structure upgrading mediates the relationship between the DE and CEI.

2.4. The Mediating Effect of Industrial Structure Advancement on the Relationship Between the DE and CEI

The mediating effect of industrial structure advancement on the relationship between the DE and CEI is primarily observed in the following three aspects.

The first is by driving the transformation towards high-end industries. The DE facilitates the flow of information and technological innovation, steering industrial structures toward high value-added, high-tech sectors (Yan et al., 2022). These advanced industries exhibit higher production efficiency and lower CEI. The evolving industrial structure increasingly favors the service sector, especially knowledge-intensive services, such as information technology, financial services, and education, which typically generate lower carbon emissions. The DE facilitates this transition by using information technology and internet platforms to boost production efficiency and minimize environmental impact. By enabling this shift through digital platforms, the DE enhances production efficiency while reducing environmental impact, thereby contributing to carbon reduction through structural upgrading.

Second is the transformation of economic growth. The advanced industrial structure promotes the growth pattern of the economy. The traditional model, reliant on resource consumption and low-value-added labor-intensive industries, is progressively being replaced by a model driven by technological innovation, increased productivity, and low-carbon development (X. Wang & Zhong, 2023). Within the framework of DE development, businesses and regions are increasingly prioritizing green productivity and sustainable practices. This shift fosters an efficient, low-carbon economic model that helps curb the rise in CEI. Additionally, the advancement of industrial structure further reduces carbon emission by transforming growth patterns.

Third is the incentivizing role of industrial policies. Industrial policy incentives play a crucial role in promoting green technology innovation and the development of low-carbon industries. Through policies such as financial subsidies, tax incentives, and innovation rewards, the government encourages enterprises to enhance their support for green, low-carbon industries and technology research and development (Gao et al., 2024). Additionally, the government uses financial instruments like green credit and green bonds to direct capital towards low-carbon industries and technologies. These mechanisms not only provide financing support for low-carbon projects but also attract investment into the green sector.

In summary, this study proposes Hypothesis (4): the advancement of industrial structure mediates the relationship between the CEI and the DE.

2.5. Threshold Effect of the DE in the Relationship Between the DE and CEI

First is the scale effect (Gao et al., 2024). In the initial stages of DE development, technological limitations and resource constraints hinder significant innovations and market effects. Consequently, the influence on reducing carbon intensity is minimal, and in a short period of time, increased economic activities may even cause a rise in carbon intensity. However, as the DE expands and technologies including the internet, big data, and artificial intelligence become more widely adopted, they stimulate large-scale green technological innovation, enhance industrial structures, and increase energy efficiency. This causes a significant reduction in carbon intensity, with scale and network effects becoming more pronounced, effectively controlling the CEI.

Second is the effect of technological accumulation. In the early stages of DE development, technological innovations primarily focused on infrastructure, such as information flow and data analysis, exert a limited direct impact on carbon reduction. However, once the DE reaches a critical threshold, accumulated technological progress facilitates the broader deployment of low-carbon technologies, energy-efficient solutions, and intelligent production methods, significantly reducing the CEI (Fan & Jia, 2024; Feng et al., 2018).

Therefore, this study proposes Hypothesis (5): there is a threshold effect in the influence of the DE on CEI.

In summary, Figure 1 outlines the mechanisms through which the DE affects CEI from direct and indirect perspectives, through the mediating effect and threshold effect. For details, see Figure 1.

3. Research Design and Method

3.1. Variables and Data Description

1.

Variables and Data Description

The dependent variable is carbon emission intensity (CEI). According to the approach of Ling et al. (2023); P. Guo and Liang (2022), the CEI was measured as the ratio of carbon dioxide emissions to a city’s Gross Domestic Product (GDP). The formula is as follows:

carbon emission intensity = carbon emissions/GDP

(1)

2.

Independent Variable

The independent variable is the Digital Economy Index (DE). Following the methodology of X. C. Zhao et al. (2023) and Xia et al. (2025), a city-level DE evaluation system was constructed based on two dimensions: digital inclusive finance and internet development. The composite value was calculated using the entropy method. Detailed information can be found in

Table 1. The city-level DE construction framework.

Variable Name	Constituent Elements	Source
Digital Economy Index	Inclusive Digital Finance Index	China City Statistical Yearbook
	Number of Internet Users per 100 People	China City Statistical Yearbook
	Proportion of Workers in Information Transmission, Computer Services, and Software Industries	China City Statistical Yearbook
	Per Capita Telecommunications Volume (CNY in ten thousand)	China City Statistical Yearbook
	Number of Mobile Phone Users per 100 People	China City Statistical Yearbook

Note: Table sourced from T. Zhao et al. (2020).

.

3.

Steps for Entropy Weight Method

(1)

Dimensionless Processing:

$$\mathrm{For} \mathrm{positive} \mathrm{indicators}\text{:} {\overline{\mathrm{x}}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}-\mathrm{m}\mathrm{i}\mathrm{n}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}}{\mathrm{m}\mathrm{a}\mathrm{x}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}-\mathrm{m}\mathrm{i}\mathrm{n}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}}}$$

(2)

$$\mathrm{For} \mathrm{negative} \mathrm{indicators}\text{:} {\overline{\mathrm{x}}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{\mathrm{m}\mathrm{a}\mathrm{x}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}{-\mathrm{X}}_{\mathrm{i}\mathrm{j}}}{\mathrm{m}\mathrm{a}\mathrm{x}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}-\mathrm{m}\mathrm{i}\mathrm{n}\left\{{\mathrm{X}}_{\mathrm{j}}\right\}}}$$

(3)

${\mathrm{X}}_{\mathrm{i}\mathrm{j}}$ is the value of the j-th indicator for the i-th sample.

${\overline{\mathrm{x}}}_{\mathrm{i}\mathrm{j}}$ means standard data processing;

(2)

Calculating the Proportion of the Indicator Value for the i-th Project under the j-th Indicator:

$$p_{ij}=r_{kj}∕{\sum }_{k=1}^n{r}_{kj}$$

(4)

(3)

Calculating the Entropy Value $e_j$ for the j-th Indicator:

$$e_j=-k{\sum }_{i=1}^m{P}_{ij} \times \mathit{ln}{p}_{ij}$$

(5)

where

$$K=1/\ln m$$

(6)

(4)

Calculating the Entropy Weight $W_j$ for the j-th Indicator:

$$W_i=(1-e_j)/{\sum }_{j=1}^n(1-e_j)$$

(7)

(5)

Calculating the Indicator Evaluation Score:

$$S_{ij} = W_i \times {\overline{x}}_{ij}$$

(8)

4.

Mediating Variable

Regarding the indicator for industrial structure upgrading (TCI), following the approach of Gan et al. (2011), the ratio of the tertiary industry’s output value to that of the secondary industry was used. TCI reflects the degree to which a city’s industrial structure has shifted toward more advanced, high-tech, and service-oriented sectors. A higher ratio indicates a more advanced industrial structure. The formula is as follows:

$${TCI}_{it}=Y_{3it}/Y_{2it}$$

(9)

$Y_{2it}$, and $Y_{3it}$ represent the output values of the secondary and tertiary industries, respectively, in region i during period t. The data was obtained from the urban statistical yearbooks of various regions.

Regarding the indicator for industrial structure upgrading (TC), this work adopted the method proposed by Ge and Zhang (2021), where weights of 1, 2, and 3 were assigned to the primary, secondary, and tertiary industries, respectively, and then summed for measurement. A higher TC value indicates a greater degree of industrial structure upgrading. The formula is as follows:

$${TC}_{it}=Y_{1it}+2 \times Y_{2it}{+3\ast Y}_{3it}$$

(10)

$Y_{1it} Y_{2it}, \mathrm{a}\mathrm{n}\mathrm{d}$ $Y_{3it}$ represent the ratios of the output values of the primary, secondary, and tertiary industries in region i during period t relative to the local GDP. The data for these calculations was taken from the regional city statistical yearbooks;

5.

Control Variable Setting

To ensure a comprehensive and robust empirical analysis, this study incorporates five control variables in addition to those previously mentioned. These include the following.

According to Xu and Zhang (2025) and Che and Han (2025), the level of government intervention (GOV) is measured by the ratio of local fiscal expenditure to GDP, reflecting the extent of governmental involvement in economic activities through resource allocation. Liang (2017) and Walenta-Bergmann and Wi (2024) define social consumption level (CON) as the ratio of total retail sales of consumer goods to GDP, which intuitively captures the proportion of consumption in overall economic output and serves as a key indicator of societal consumption. Nam and Ryu (2025) assess the level of economic development (PGDP) using the natural logarithm of per capita GDP. This approach mitigates heteroscedasticity, ensuring more stable and comparable data for evaluating regional economic performance. Wijethunga et al. (2025) measure the financial development level (FDL) by the ratio of year-end loans and deposits of financial institutions to GDP, offering a comprehensive view of financial resource allocation and market activity. Lin et al. (2023) evaluate educational level (EDU) through the ratio of education expenditure to total local fiscal expenditure, reflecting the government’s prioritization of education within its overall fiscal strategy. All variable data used in this study are sourced from the China City Statistical Yearbook and the WIND database;

6.

Data Description and Descriptive Statistical Analysis

Given that the concept of the DE was formally introduced in China only in 2017 and acknowledging substantial data limitations for municipal-level cities prior to 2010, this study excludes certain cities with incomplete records to ensure the accuracy and reliability of the empirical analysis. Consequently, the final research sample comprises panel data from 274 prefecture-level cities in China, spanning the period from 2011 to 2022. Missing values are addressed using linear interpolation.

Table 2. Descriptive statistics of variables.

Variable	N	Mean	SD	Min	Max
CEI	3288	0.099	0.092	0.004	0.892
DE	3288	0.345	0.333	0.006	2.885
TC	3288	2.310	0.144	1.821	2.835
TCI	3288	1.081	0.611	0.175	5.650
GOV	3288	0.201	0.101	0.044	0.916
CON	3288	0.384	0.109	0.001	1.013
FDL	3288	2.576	1.232	0.587	21.302
PGDP	3288	16.685	0.949	14.106	19.917
EDU	3288	0.176	0.039	0.036	0.356

Note: Self-organized findings.

presents the descriptive statistics for each variable. Some variables display significant disparities, particularly in the substantial gap between maximum and minimum values. This suggests considerable regional variation in the factors represented by these variables.

7.

Weight matrix involved in this paper

(1)

Spatial distance matrix and calculation formula:

$$W_d=\left\{\begin{array}{c}{\displaystyle \frac{1}{d_{ij}}},i\ne j\\ 0, i=j\end{array}\right.$$

(11)

where d is computed using the latitude and longitude coordinates of each prefecture-level city;

(2)

Adjacency matrix:

$$W_{ij}=\left\{\begin{array}{c}1, \mathrm{region} i \mathrm{is} \mathrm{adjacent} \mathrm{to} \mathrm{region} j.\\ 0, \mathrm{region} i \mathrm{and} \mathrm{region} j \mathrm{are} \mathrm{not} \mathrm{adjacent};\end{array}\right.$$

(12)

(3)

Economic geography nesting matrix:

$$W= {mW}_1 +(1-m)W_2$$

(13)

$$W_1 =1/d_{ij}$$

(14)

$$W_2 = E_i{E}_j \times e^{-\beta d_{ij}}$$

(15)

$W_2$ is the economic distance matrix, where β = 100; $d_{ij}$ denotes the distance between regions i and j, measured in miles; and m denotes the weight of the geographic distance matrix, with m ranging between 0 and 1;

(4)

Economic distance spatial weight matrix:

$$W=W_d\mathit{diag}(\mathit{pGD}{P}_i-\mathit{pGD}{P}_i)$$

(16)

W_d represents the geographic matrix, and pGDP denotes the per capita gross domestic product.

3.2. Spatial Autocorrelation Test of the CEI

Table 3. Global Moran and Geary indices of CEI under the spatial economic distance matrix weighting matrix.

Year	Moran’s I	p-Value	Geary’s c	p-Value
2011	0.331	0.0000	0.204	0.025
2012	0.297	0.0000	0.223	0.015
2013	0.317	0.0000	0.211	0.016
2014	0.329	0.0000	0.213	0.0165
2015	0.337	0.0000	0.199	0.005
2016	0.371	0.0000	0.165	0.003
2017	0.320	0.0000	0.270	0.002
2018	0.319	0.0000	0.226	0.003
2019	0.348	0.0000	0.243	0.003
2020	0.348	0.0000	0.245	0.001
2021	0.332	0.0000	0.272	0.003
2022	0.449	0.0000	0.177	0.000

Note: Self-organized findings.

presents the global Moran’s I index and Geary’s C index for CEI, calculated using the spatial economic–geographic distance weight matrix. The results indicate that the Global Moran’s I index is positive and statistically significant at the 1% level. Over time, the Moran’s I value increased steadily from 0.331 in 2012 to 0.449 in 2022, indicating a growing spatial clustering effect of CEI. This confirms the presence of spatial autocorrelation in CEI. Additionally, the global Geary’s C index is less than one, also passing the 1% significance level. The analysis of the Table 3 indices reveals that carbon emission intensity exhibits spatial clustering, with high–high and low–low associations. To put it simply, regions with high carbon emission intensity tend to cluster with other high-intensity regions, while areas with low CEI are more likely to be adjacent to similarly low-intensity regions.

Using Stata 17.0 software, Local Moran scatter plots and LISA (Local Indicators of Spatial Association) maps of carbon emission intensity are generated for the years 2012, 2015, 2018, and 2022. Analysis of Figure 2, Figure 3, Figure 4 and Figure 5 reveals that the most of the observations are concentrated in the first and third quadrants. The observations in the first quadrant exhibit a “high–high clustering” pattern, indicating that regions with high CEI are adjacent to one another. In contrast, the observations in the third quadrant display a “low–low clustering” pattern, meaning that regions with low carbon emission intensity levels are situated next to each other. Additionally, a small number of regions have Moran values that fall in the second and fourth quadrants. This phenomenon demonstrates that there are significant spatial disparities in CEI levels across different prefecture-level cities. Simultaneously, LISA maps for the aforementioned four years were generated using ARGICS 10.8 software. The LISA maps for certain years provide a more intuitive observation of the distribution of H–H (high-value clustering) and L–L (low-value clustering) patterns. Among these, the H–H and L–H (low-value surrounded by high-value) patterns are primarily concentrated in the eastern regions, while the L–L and H–L (high-value surrounded by low-value) patterns are mainly found in the central regions, exhibiting a distinct regional distribution pattern. This distribution disparity is likely linked to regional economic development models, industrial structures, and energy consumption patterns, warranting further in-depth investigation.

3.3. Spatial Model Selection

Table 4. LM, LR, and Husman test.

Numerical Value		Numerical Value
LM-error	1420.471 ***	R-LM-error	1225.786 ***
LM-lag	194.761 ***	R-LM-lag	0.076
LR-lrtest sdm_a sar_a	55.52 ***	LR-lrtest sdm_a sem_a	83.15 ***
wald-sdm		21.4 ***
Hausman		135.22 ***

In the table, *** represent significance at the 1% levels, respectively. Numerical values represent standard errors.

reports the relevance test results for the spatial economic–geographic weight matrix. Both the Lagrange Multiplier (LM) test for spatial error and the Robust LM (R-LM) test yield statistically significant results, indicating that spatial error dependence should be accounted for in model specification. However, while the R-LM test for spatial lag shows some performance, it does not pass the test of significance. This preliminarily suggests that the spatial lag model has certain limitations. Overall, both the SEM and the SLM are considered in the preliminary selection. Furthermore, the LR-SDM test results passe the significance test, and the Wald-SDM is also statistically significant., strongly indicating that the SDM performs well in fitting the data. Additionally, The Hausman test value is 135.22, indicating statistical significance, providing stronger support for the model selection.

In conclusion, after a series of rigorous tests and analyses, the SDM with double fixed effects is the most suitable choice. This model provides a more holistic and precise representation of spatial relationships among variables and their effects.

3.4. Model Specification

3.4.1. Benchmark Model Specification

In the direct perspective analysis, our study uses a basic regression model to examine the impact of the DE on CEI. This paper refers to the model design by J. Li (2024), as follows:

$${CEI}_{it}={\beta }_0+{{\beta }_{1}DE}_{it}+{\gamma Controls+v}_{it}+{\epsilon }_{it}$$

(17)

Here, β₀ represents the constant, i denotes the prefecture-level city, t represents the year, and CEI stands for carbon emission intensity; DE refers to the digital economy level, Controls represent the control variables, ε is the error term, u denotes the time fixed effects, and v represents the individual fixed effects.

3.4.2. Spatial Durbin Model Specification

After analyzing the spatial model selection, it was found that the selection of a double-fixed SDM was the most appropriate for the indirect perspective analysis in this paper. Thus, this paper refers to the spatial model design by Chen and Zhang (2024), as follows:

$${CEI}_{it}=pW{ \times CEI}_{it}+{{\beta }_{1}DE}_{it}+{{\beta }_{2}W\times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(18)

Here, w represents the spatial weight matrix. This study employs the spatial economic distance matrix for empirical analysis. u denotes the time fixed effects, v denotes the individual fixed effects, and all other terms remain identical to those in (18).

3.4.3. The Two-Region SDM Design

This study adopts a more refined approach by accounting for the heterogeneity between resource-based and non-resource-based cities, as well as the varying spatial dependence effects across regions, while integrating all variables into a single model. Drawing from Xing et al. (2023) and referencing the “National Sustainable Development Plan for Resource-based Cities (2013–2020),” we categorized 274 prefecture-level cities into 107 resource-based and 167 non-resource-based cities. In accordance with the SDM design, and following Elhorst and Frére (2009), the two-region Durbin model was formulated as follows:

$${\mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i}\mathrm{t}}={\mathrm{p}}_1{\mathrm{d}}_{\mathrm{i}\mathrm{t}}\mathrm{W}{ \times \mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i}\mathrm{t}}+{\mathrm{p}}_2{(1-\mathrm{d}}_{\mathrm{i}\mathrm{t}})\mathrm{W}{ \times \mathrm{C}\mathrm{E}\mathrm{I}}_{\mathrm{i}\mathrm{t}}+{{\mathsf{\beta }}_1\mathrm{D}\mathrm{E}}_{\mathrm{i}\mathrm{t}}+{{\mathsf{\beta }}_2\mathrm{W}\times \mathrm{D}\mathrm{E}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\gamma }\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}+{\mathrm{V}}_{\mathrm{i}\mathrm{t}}+{\mathrm{u}}_{\mathrm{i}}$$

(19)

In this model, ${\mathrm{p}}_1, {\mathrm{p}}_2{, \mathsf{\beta }}_1$, and ${\mathsf{\beta }}_2$ represent the estimated parameters and ${\mathrm{d}}_{\mathrm{i}\mathrm{t}}$ is a dummy variable, where resource-based cities are assigned a value of 1 and non-resource-based cities a value of 0. All other settings are consistent with the model specification in (18).

3.4.4. Mediation Effects Model

To further examine whether industrial advancement and upgrading mediate the relationship between the DE and CEI in prefecture-level cities, and to assess their effects, this study conducted a mediation analysis to quantify effects. Following the stepwise regression method presented by J. Wang and Guo (2023), the mediating effect model in the panel data SDM was formulated as follows:

$${CEI}_{it}={\beta }_0+p_{1}W{ \times CEI}_{it}+{{\beta }_{1}DE}_{it}+{{\beta }_{2}W \times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(20)

$${TI}_{it}={{\alpha }_0+p_{2}W \times TI}_{it}{{+\alpha }_{1}DE}_{it}+{{\alpha }_{2}W \times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(21)

$${CEI}_{it}={\gamma }_0+p_{3}W{ \times CEI}_{it}+{{\gamma }_{1}TI}_{it}{+{\gamma }_{2}W \times TI}_{it}{{+\gamma }_{3}DE}_{it}+{{\gamma }_{4}W \times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(22)

$${TCI}_{it}={{\alpha }_0+p_{4}W \times TCI}_{it}{{+\alpha }_{1}DE}_{it}+{{\alpha }_{2}W \times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(23)

$${CEI}_{it}={\gamma }_0+p_{5}W{ \times CEI}_{it}+{{\gamma }_{1}TCI}_{it}{+{\gamma }_{2}W \times TCI}_{it}{{+\gamma }_{3}DE}_{it}+{{\gamma }_{4}W \times DE}_{it}+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(24)

In this model, $p_1, p_2, p_3,p_4$, and $p_5$ represent the spatial lag coefficients of the dependent variables, TI stands for the overall industrial structure upgrading, and TCI represents the advancement of the industrial structure. ${\beta }_0$, ${\mathsf{\alpha }}_0$, ${\gamma }_0$, ${\mathsf{\delta }}_0$, and ${\theta }_0$ are the estimated constants, while ${\beta }_1, {\beta }_2, {\alpha }_1, {\alpha }_2, {\delta }_1, {\delta }_2, {\gamma }_1, {\gamma }_2, {\gamma }_3, {\gamma }_4$, ${\theta }_1, {\theta }_2$, ${\theta }_3$, and ${\theta }_4$ are the model parameters to be estimated. All other aspects are consistent with (18).

3.4.5. Threshold Model Design

To further investigate whether a threshold effect exists for the DE, this paper adopts Hansen’s (1999) threshold regression model and incorporates the DE as a threshold variable for analysis. Stata17 was applied to estimate the model through the self-sampling method (Bootstrap) with 500 iterations under single, double, and triple threshold settings, respectively.

$${CEI}_{it}={\beta }_0+{{\beta }_{1}DE}_{it} \times I({DE}_{it}\le {\chi }_1)+{\beta }_2{DE}_{it} \times I({{\chi }_1<DE}_{it}<{\chi }_2)+{\beta }_3{DE}_{it} \times I({DE}_{it}\ge {\chi }_2)+{\gamma Controls}_{it}+{\epsilon }_{it}+V_{it}+u_i$$

(25)

The meanings of the letters are the same as in (18).

4. Empirical Results Analysis and Discussion

Prior to model estimation, a multicollinearity test was performed using the Variance Inflation Factor (VIF) to avoid potential spurious regression.

Table 5. Multicollinearity regression.

Variables	VIF	1/VIF
CEI	2.24	0.447308
GOV	2.24	0.4473
CON	1.86	0.5368
FDL	1.46	0.6858
PGDP2	2.85	0.3510
EDU	1.13	0.8848
Mean VIF	1.78

Note: Self-organized findings.

shows that all variables have VIF values below 10. As a general guideline, a VIF below five suggests that multicollinearity is not a significant issue (Vittinghoff et al., 2012). This favorable result ensures the stability of the model’s estimates and supports the accuracy of subsequent predictions, enhancing the reliability of the empirical analysis.

4.1. Analysis of Basic Regression Results

Table 6. Basic regression results.

Variables	(1)	(2)
	CEI	CEI
DE	−0.038 *** (−7.88)	−0.114 *** (−5.44)
GOV		−0.070 (−0.90)
CON		0.229 *** (9.16)
FDL		−0.025 *** (−3.36)
PGDP2		−0.123 *** (−9.93)
EDU		−0.456 *** (−3.33)
Constant	0.112 *** (48.79)	2.269 *** (9.87)
Individual Fixed	NO	NO
Time Fixed	NO	NO
Observations	3288	3288
Number of id	274	274
R-squared	0.019	0.315

*** represent significance at the 1% levels, respectively; the values represent standard errors.

presents the baseline regression results of the DE’s effect on CEI. The regression results in Table 6, Column (1) indicate that the DE significantly reduces CEI. Table 6, Column (2) presents the regression results after incorporating control variables, confirming that the DE’s suppressive effect on CEI remains significant. Thus, Hypothesis (1) is confirmed.

4.2. Analysis of Spatial Empirical Results

4.2.1. Analysis of Spatial Durbin Regression Results

Table 7. Spatial economic distance matrix SDM regression results.

Variables	(1)	(2)	(3)	(4)
	CEI	CEI	CEI	CEI
DE	−0.030 *** (−3.02)	−0.123 *** (−4.90)
GOV	−0.069 ** (−2.37)	−0.269 *** (−3.75)
CON	0.013 (1.17)	0.068 ** (2.55)
FDL	−0.003 * (−1.86)	−0.006 * (−1.67)
PGDP	−0.025 *** (−3.64)	−0.054 *** (−3.32)
EDU	−0.206 *** (−5.06)	−0.443 *** (−4.05)
rho			0.635 *** (25.06)
sigma2_e				0.001 *** (39.72)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288
R-squared	0.001	0.001	0.001	0.001
Number of id	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

(2) reports the SDM regression results based on the spatial economic distance matrix. Result (3) reports a rho value of 0.635, which is statistically significant at the 1% level, confirming strong spatial spillover effects. This means that changes in CEI in one region influence surrounding regions. Result (1) shows that the digital economy (DE) significantly reduces CEI, with a coefficient of –0.03. Holding other factors constant, a 1% increase in the DE is associated with a 0.03% decrease in CEI, confirming the DE’s effectiveness in mitigating carbon emissions.

4.2.2. Spatial Durbin Decomposition Regression

To better capture the spatial impact of the DE on CEI, both direct and indirect effects, this study uses the maximum likelihood (ML) method proposed by Elhorst (2014) to decompose the SDM.

Table 8. SDM regression decomposition results.

Variables	CEI
	Direct Effect	Indirect Effect	Total Effect
DE	−0.044 *** (−4.39)	−0.374 *** (−5.67)	−0.419 *** (−6.10)
GOV	−0.104 *** (−3.70)	−0.825 *** (−4.54)	−0.930 *** (−4.94)
CON	0.023 ** (2.08)	0.207 *** (3.049)	0.230 *** (3.24)
FDL	−0.003 ** (−2.29)	−0.019 ** (−2.05)	−0.022 ** (−2.23)
PGDP	−0.0325 *** (−4.78)	−0.183 *** (−4.45)	−0.216 *** (−5.11)
EDU	−0.2671 *** (−6.31)	−1.490 *** (−5.08)	−1.757 *** (−5.66)
rho	0.635 *** (25.057)
sigma2_e	0.0012 *** (39.722)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288
R-squared	0.001	0.001	0.001
Number of id	274	274	274

**, and *** represent significance at the 5%, and 1% levels, respectively; the values represent standard errors.

presents the regression results, showing that all three effects of the DE significantly reduce CEI. According to the foregoing analysis, Hypothesis (2) is supported. The absolute value of the indirect effect coefficient (0.374) is greater than that of the direct effect coefficient (0.044), likely because the DE generates spillover effects through multiple channels, including technology, industry, policy, consumption, and markets. These spillover effects tend to exert a stronger effect on neighboring regions.

4.2.3. The Decay Boundary of Spatial Spillover Effects

To better understand the spatial spillover effects of the DE on CEI across varying regional distances, Figure 6 presents regressions based on the SDM and spatial distance matrix, with intervals of 100 km. It plots the spatial spillover coefficients of the DE on CEI along with their 90% confidence intervals. The distance thresholds in Figure 6 can be divided into three segments: (1) when the distance threshold is less than 900 km, the DE significantly suppresses CEI in neighboring regions. Within shorter distances, technological innovations, digital platforms, and information flow driven by the DE can typically enhance carbon emission efficiency and the adoption of green technologies through spatial diffusion effects. Neighboring cities often benefit from the innovations of the central digital economy city, leveraging shared knowledge, technologies, and experiences to improve local green productivity and energy efficiency, thereby reducing CEI. Enterprises in neighboring cities may emulate the low-carbon development model of the central city, adopting energy-saving and emission-reduction technologies, and green production processes, etc., thus driving a reduction in local CEI. As the threshold moves below 300 km, the suppressive effect weakens due to increased geographical distance, which raises the costs of cross-regional cooperation and communication. As a result, the spillover effects of knowledge and technology weaken over time. Furthermore, innovations from developed regions may not be absorbed effectively by underdeveloped areas due to geographical and economic differences. (2) When the distance threshold is between 900 km and 1300 km, the DE still exerts a suppressive effect on carbon emission spillover, though the effect is no longer significant. The reason for this may be due to the rising costs of cross-regional cooperation and communication as distance increases. While knowledge and technology can be shared through face-to-face communication and cooperation between neighboring cities, these effects weaken with greater distances due to factors like transportation costs, delayed information dissemination, and lack of cooperative relationships. Additionally, the limited resources and capabilities of neighboring regions may restrict the effective absorption of spillover technologies. (3) When the distance threshold is between 1300 km and 2000 km, the spatial spillover effect of the DE on CEI significantly weakens, but it still exerts a suppressive effect.

In conclusion, the spatial spillover effect of the DE on CEI shows a diminishing trend, from suppression to no effect, supporting the hypothesis of spatial decay effects. This is consistent with Rogers’ (2003) view that innovations tend to diffuse more effectively within socially and spatially connected clusters, but the effect diminishes with distance.

4.3. Robustness Check

To strengthen the robustness of the SDM regression on CEI in relation to the DE and ensure that the results remain unaffected by changes in other variables, a robustness check was performed. This paper presents the SDM robustness regression based on the spatial geographic weight matrix, as well as SDM regressions conducted by changing the weight matrix, using the economic geography nested matrix and zero-order matrix, and excluding provincial capitals and sub-provincial cities. The specific results can be found in

Table 9. SDM robustness regression results with economic geography distance weights.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.030 ** (−2.30)	−0.123 *** (−3.46)			−0.045 *** (−2.95)	−0.374 *** (−3.29)	−0.419 *** (−3.42)
GOV	−0.069 (−1.46)	−0.269 *** (−3.00)			−0.104 ** (−2.11)	−0.829 *** (−2.86)	−0.933 *** (−2.97)
CON	0.013 (0.87)	0.068 (1.51)			0.024 (1.55)	0.216 * (1.74)	0.239 * (1.85)
FDL	−0.003 (−0.85)	−0.006 * (−1.70)			−0.003 (−1.06)	−0.020 (−1.61)	−0.023 (−1.58)
PGDP	−0.025 (−1.39)	−0.054 ** (−2.14)			−0.032 * (−1.66)	−0.186 ** (−2.23)	−0.218 ** (−2.32)
EDU	−0.206 * (−1.71)	−0.443 *** (−3.31)			−0.266 ** (−2.04)	−1.489 *** (−2.90)	−1.754 *** (−2.91)
rho			0.635 *** (15.64)
sigma2_e				0.001 *** (4.47)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.001	0.001	0.001	0.001	0.001	0.001	0.001
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

,

Table 10. Regression results with economic geography nested matrix.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.040 *** (−4.12)	−0.128 *** (−4.60)			−0.052 *** (−4.99)	−0.307 *** (−5.13)	−0.359 *** (−5.64)
GOV	−0.118 *** (−4.10)	−0.146 * (−1.92)			−0.136 *** (−4.78)	−0.436 *** (−2.83)	−0.572 *** (−3.51)
CON	0.024 ** (2.12)	0.060 * (1.91)			0.031 *** (2.71)	0.157 ** (2.40)	0.188 *** (2.67)
FDL	−0.004 ** (−2.45)	0.001 (0.20)			−0.004 ** (−2.43)	−0.002 (−0.24)	−0.006 (−0.64)
PGDP	−0.031 *** (−4.80)	−0.060 *** (−3.52)			−0.038 *** (−5.50)	−0.159 *** (−4.33)	−0.197 *** (−5.00)
EDU	−0.249 *** (−5.92)	−0.594 *** (−4.79)			−0.307 *** (−7.12)	−1.481 *** (−5.63)	−1.788 *** (−6.37)
rho			0.533 *** (18.80)
sigma2_e				0.001 *** (39.86)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.005	0.005	0.005	0.005	0.005	0.005	0.005
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

,

Table 11. Regression results with adjacency matrix.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.067 *** (−6.55)	−0.014 (−0.90)			−0.072 *** (−6.81)	−0.077 *** (−3.02)	−0.149 *** (−5.14)
GOV	−0.062 ** (−2.13)	−0.146 *** (−3.18)			−0.086 *** (−3.02)	−0.298 *** (−4.09)	−0.383 *** (−4.64)
CON	0.020 (1.51)	−0.034 * (−1.74)			0.018 (1.48)	−0.040 (−1.41)	−0.022 (−0.74)
FDL	−0.000 (−0.03)	−0.003 (−1.03)			−0.000 (−0.28)	−0.005 (−1.04)	−0.005 (−0.99)
PGDP	−0.063 *** (−7.34)	0.027 *** (2.77)			−0.063 *** (−7.36)	−0.003 (−0.24)	−0.066 *** (−4.18)
EDU	−0.301 *** (−7.06)	−0.053 (−0.73)			−0.324 *** (−7.59)	−0.322 *** (−2.63)	−0.646 *** (−4.64)
rho			0.464 *** (22.31)
sigma2_e				0.001 *** (39.32)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.004	0.004	0.004	0.004	0.004	0.004	0.004
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

and

Table 12. Regression results excluding sub-provincial cities and provincial capitals.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.018 ** (−2.37)	−0.066 *** (−3.32)			−0.025 *** (−3.13)	−0.165 *** (−3.90)	−0.190 *** (−4.29)
GOV	−0.055 ** (−2.40)	−0.092 (−1.55)			−0.067 *** (−3.06)	−0.266 ** (−2.20)	−0.333 *** (−2.67)
CON	0.014 (1.58)	−0.040 * (−1.75)			0.012 (1.42)	−0.066 (−1.41)	−0.054 (−1.10)
FDL	0.000 (0.15)	−0.001 (−0.31)			0.000 (0.07)	−0.002 (−0.26)	−0.002 (−0.23)
PGDP	−0.008 (−1.37)	−0.006 (−0.46)			−0.009 (−1.57)	−0.024 (−0.85)	−0.032 (−1.15)
EDU	−0.066 ** (−2.03)	−0.292 *** (−3.26)			−0.095 *** (−2.83)	−0.697 *** (−3.56)	−0.792 *** (−3.79)
rho			0.555 *** (18.42)
sigma2_e				0.001 *** (37.21)
Individual Fixed	YES
Time Fixed	YES
Observations	2880	2880	2880	2880	2880	2880	2880
R-squared	0.044	0.044	0.044	0.044	0.044	0.044	0.044
Number of id	240	240	240	240	240	240	240

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

. The spatial spillover coefficient rho in each table passes the significance test. The core explanatory variable, the DE, has a significantly negative impact on the direct, indirect, and total effects, indicating that it effectively reduces the CEI across regions. The results across all tables are consistent with those from the previous regressions, indicating that the empirical findings remain stable despite variations in the weight matrix and the exclusion of certain cities. These findings withstand the robustness check, confirming the reliability of the results.

4.4. Endogeneity Test

In the model design involving the DE and CEI, reverse causality or endogeneity between the explanatory variable and the error term (Wooldridge, 2006), as well as the omission of important relevant variables, may lead to biased model estimates. To address this issue, this study adopts the methods of Sheng et al. (2023) and Chen and Jiang (2023), utilizing the lagged DE and the product of the distance from each region to Hangzhou as instrumental variables. The 2SLS and system GMM methods were employed to test for endogeneity.

Table 13. 2SLS and system GMM test results.

Variables	2sls	System-GMM
	CEI	CEI
L. CEI		0.635 *** (33.62)
DE	−0.110 *** (−4.13)	−0.120 *** (−8.10)
GOV	−0.135 *** (−3.56)	−0.292 *** (−7.17)
CON	0.234 *** (15.40)	0.253 *** (15.07)
FDL	−0.025 *** (−13.09)	−0.015 *** (−7.32)
PGDP	−0.141 *** (−20.55)	−0.119 *** (−17.16)
EDU	−0.593 *** (−9.27)	−0.588 *** (−7.89)
Constant		2.160 *** (17.27)
Kleibergen–Paap LM statistic	945.768
p-value	0.0000
Cragg–Donald Wald F statistic	1441.134
AR(1)z and p-value		−6.6454 0.0000
AR(2)z and p-value		−0.84315 0.3991
Sargan test		172 0.5423
Number of id	274	274
R-squared	0.325

*** represent significance at the 1% levels, respectively; the values represent standard errors.

reports the results from both methods. In both cases, the core explanatory variable, the DE, exhibits a significant suppressive effect on CEI. Under the 2SLS method, the LM statistic is 945.768 and is statistically significant, confirming the validity of the identification test. The F-statistic is 1441.134, which is much greater than 19, indicating that weak instrument issues are not present. Both tests confirm the reliability of the model. In the system GMM test, AR(1) is statistically significant, while AR(2) is not, and the Sargan test fails as well. Based on the results of these three tests, the model passes the system GMM test, confirming its robustness.

4.5. Heterogeneity Analysis

Given the significant regional disparities in both the DE and CEI across China, t is necessary to perform a region-specific analysis to better capture the spatial heterogeneity in their relationship. This study follows the classification of Dou and Guan (2023) for spatial regional heterogeneity analysis. The 274 prefecture-level cities in China examined in this study were categorized into 99 eastern, 98 central, and 77 western cities based on geographical location. SDM regression analysis was then conducted for each group.

4.5.1. Eastern Cities

Table 14. SDM regression results for eastern cities under the spatial economic distance matrix.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.138 *** (−3.05)	−0.403 *** (−3.94)			−0.197 *** (−3.92)	−0.969 *** (−4.27)	−1.166 *** (−4.63)
GOV	0.133 (1.43)	−0.403 ** (−2.36)			0.087 (0.98)	−0.679 ** (−2.10)	−0.592 * (−1.75)
CON	−0.079 *** (−3.22)	0.017 (0.42)			−0.080 *** (−3.39)	−0.045 (−0.61)	−0.125 (−1.56)
FDL	−0.011 *** (−2.68)	−0.012 * (−1.81)			−0.013 *** (−3.11)	−0.035 ** (−2.51)	−0.048 *** (−2.99)
PGDP	−0.057 *** (−3.35)	−0.190 *** (−5.66)			−0.085 *** (−4.95)	−0.448 *** (−6.39)	−0.533 *** (−7.24)
EDU	−0.424 *** (−4.30)	−0.223 (−1.11)			−0.477 *** (−4.80)	−0.894 ** (−2.19)	−1.371 *** (−3.07)
rho			0.536 *** (13.65)
sigma2_e				0.002 *** (23.61)
Individual Fixed	YES
Time Fixed	YES
Observations	1188	1188	1188	1188	1188	1188	1188
R-squared	0.035	0.035	0.035	0.035	0.035	0.035	0.035
Number of id	99	99	99	99	99	99	99

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

reports the results for the 100 eastern cities. In Table 14, Column (3) the rho value is 0.536 and is significantly positive, indicating spatial spillover effects of the DE on CEI in 100 eastern cities. In Table 13, Column (1), the DE also exerts a negative effect on the CEI of the region. The indirect effect is notably large and negative (−0.969), indicating that through spatial spillover effects, changes in carbon emissions significantly affect neighboring cities, with this impact being negatively correlated with CEI in these regions.

4.5.2. Central Cities

Table 15. Results of SDM central city regression under spatial economic distance matrix.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	0.002 (0.14)	−0.010 (−0.41)			0.001 (0.12)	−0.015 (−0.40)	−0.013 (−0.33)
GOV	−0.069 ** (−2.20)	0.020 (0.28)			−0.071 ** (−2.37)	−0.012 (−0.11)	−0.082 (−0.76)
CON	0.021 * (1.77)	−0.011 (−0.41)			0.022 ** (2.02)	−0.002 (−0.05)	0.020 (0.53)
FDL	−0.000 (−0.31)	−0.000 (−0.11)			−0.000 (−0.34)	−0.001 (−0.13)	−0.001 (−0.20)
PGDP	−0.006 (−0.75)	0.037 ** (2.39)			−0.003 (−0.45)	0.053 ** (2.29)	0.050 ** (2.16)
EDU	−0.095 ** (−2.26)	−0.166 * (−1.66)			−0.106 *** (−2.58)	−0.306 ** (−1.99)	−0.412 ** (−2.49)
rho			0.382 *** (7.88)
sigma2_e				0.000 *** (23.90)
Individual Fixed	YES
Time Fixed	YES
Observations	1176	1176	1176	1176	1176	1176	1176
R-squared	0.015	0.015	0.015	0.015	0.015	0.015	0.015
Number of id	98	98	98	98	98	98	98

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

reports the regression results for central cities, revealing a spatial lag coefficient (rho) of 0.382, which is statistically significant, confirming the existence of spatial spillover effects. However, the DE is not statistically significant. This suggests that while the DE has the potential to reduce carbon intensity in the region, its overall effect is inhibitory. The industrial structure of central cities being overly reliant on high-carbon industries and the continued dependence on traditional fossil fuels limits the DE’s ability to drive a green transition and significantly reduce carbon emissions in the short term. Additionally, local governments in central regions may prioritize short-term economic growth over sustainable development.

4.5.3. Western Cities

According to

Table 16. Results of SDM western cities regression under spatial economic distance matrix.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
DE	−0.032 *** (−3.13)	−0.056 ** (−2.39)			−0.035 *** (−3.33)	−0.081 *** (−2.74)	−0.116 *** (−3.84)
GOV	−0.054 (−1.52)	−0.303 *** (−3.45)			−0.069 ** (−2.03)	−0.403 *** (−3.76)	−0.472 *** (−4.17)
CON	0.075 *** (3.74)	0.031 (0.69)			0.079 *** (4.10)	0.068 (1.17)	0.146 ** (2.33)
FDL	0.010 ** (2.44)	0.026 *** (2.93)			0.011 *** (2.78)	0.036 *** (3.29)	0.047 *** (3.79)
PGDP	−0.004 (−0.39)	0.008 (0.33)			−0.004 (−0.37)	0.008 (0.27)	0.004 (0.14)
EDU	−0.154 *** (−2.79)	−0.133 (−1.00)			−0.159 *** (−2.96)	−0.204 (−1.20)	−0.364 ** (−1.98)
rho			0.238 *** (4.55)
sigma2_e				0.001 *** (21.36)
Individual Fixed	YES
Time Fixed	YES
Observations	924	924	924	924	924	924	924
R-squared	0.061	0.061	0.061	0.061	0.061	0.061	0.061
Number of id	77	77	77	77	77	77	77

**, and *** represent significance at the 5%, and 1% levels, respectively; the values represent standard errors.

, the DE has a negative and statistically significant direct effect on CEI (−0.032) in the region. This indicates that a higher DE level in the region reduces CEI. The indirect effect is also negative (−0.081), suggesting that carbon emissions exert a significant inhibitory effect on neighboring cities through spatial spillovers, reducing carbon intensity in those areas. The total effect, encompassing both direct and indirect effects, is negative (−0.116), further reinforcing the overall reduction in CEI.

4.6. Further Differentiation Between Resource and Non-Resource Cities

Table 17. The results of the two-region SDM regression.

Variables	Economic Geography Matrix Results	Economic Geography Nested Matrix Results
ρ1	0.787 *** (32.710)	3.362 *** (81.205)
ρ2	0.154 *** (3.579)	1.192 *** (20.363)
ρ1 − ρ2	0.634 *** (13.323)	2.170 *** (28.889)
DE	0.019 ** (4.316)	0.022 ** (5.244)
GOV	−0.003 (−0.153)	−0.089 *** (−6.910)
CON	−0.021 *** (−2.054)	−0.059 *** (−5.813)
FDL	0.019 *** (19.274)	0.023 *** (24.009)
PGDP	0.017 *** (8.581)	0.0029 *** (16.530)
EDU	−0.087 (0.5928)	−0.492 *** (16.069)
con	−0.016 *** (−7.116)	−0.004 * (−1.777)
w × DE	0.005 (0.388)	−0.087 * (−1.867)
w × GOV	−0.084 *** (−2.156)	0.631 *** (3.566)
w × CON	−0.005 (−0.175)	0.096 (1.226)
w × FDL	−0.008 *** (−3.548)	−0.047 *** (−5.453)
w × PGDP	0.015 *** (3.295)	−0.015 (−0.798)
w × EDU	0.0004 (0.0007)	0.955 *** (5.252)
Individual Fixed	NO
Time Fixed	YES
R2	0.6476	0.7033
Observations	3288
Number of id	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

presents the results of the two-region SDM regression, implemented in MATLAB R2017, using both the spatial economic distance weight matrix and the economic-geography nested matrix. The results in the table show that p₁ and p₂ represent the spatial spillover correlation coefficients for resource-based cities and non-resource cities, respectively. In the spatial economic distance weight matrix, the values for resource-based cities (0.787) and non-central cities (0.154) both pass the significance test. In the economic–geography nested matrix, the values for central cities (3.362) and non-central cities (1.192) are also statistically significant. By comparing the difference between p₁ and p₂, which are 0.634 and 2.170, respectively, both pass the 5% significance level. This indicates that the spillover effect of the DE on CEI is significantly greater in resource-based cities than in non-resource-based cities. This suggests that resource-based cities face greater pressure for industrial transformation, and the DE can effectively drive technological innovation, green development, and industrial upgrading, leading to a stronger spillover effect. In contrast, non-resource-based cities tend to have more diversified and low-carbon-oriented economic structures, thereby limiting the marginal impact and spillover potential of the DE on regional carbon reduction.

4.7. Analysis of Intermediary Pathway Results

The regression results of Equation (20) are reported in Table 8. As shown in Table 8, the DE significantly suppresses CEI through direct, indirect, and total effects. The results of Equation (21) are reported in

Table 18. TI spatial Durbin regression results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	TI	TI	TI	TI	Direct Effect	Indirect Effect	Total Effect
DE	0.069 *** (6.43)	−0.150 *** (−5.47)			0.065 *** (5.98)	−0.172 *** (−4.85)	−0.107 *** (−3.00)
GOV	0.022 (0.69)	−0.273 *** (−3.48)			0.011 (0.38)	−0.348 *** (−3.51)	−0.336 *** (−3.41)
CON	0.182 *** (14.68)	0.018 (0.61)			0.185 *** (15.87)	0.084 ** (2.30)	0.269 *** (7.32)
FDL	0.006 *** (3.59)	0.003 (0.71)			0.006 *** (3.75)	0.005 (1.12)	0.011 ** (2.11)
PGDP2	0.032 *** (4.28)	−0.037 ** (−2.07)			0.031 *** (4.21)	−0.038 * (−1.69)	−0.007 (−0.31)
EDU	0.113 ** (2.53)	−0.385 *** (−3.23)			0.102 ** (2.35)	−0.454 *** (−2.89)	−0.352 ** (−2.12)
rho			0.248 *** (7.88)
sigma2_e				0.002 *** (40.34)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.456	0.456	0.456	0.456	0.456	0.456	0.456
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

. From the regression results, it can be observed that the DE exerts a suppressive effect on the overall industrial structure upgrading in terms of total utility, indicating that there are difficulties in the transformation of domestic high-carbon and traditional industries. The total effect coefficient of −0.107 in Table 18, Column (7) and the total effect coefficient of 0.526 in

Table 19. CEI and TI spatial Durbin regression results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
TI	0.092 *** (5.78)	0.111 *** (2.70)			0.110 *** (6.53)	0.415 *** (4.31)	0.526 *** (5.17)
DE	−0.035 *** (−3.59)	−0.113 *** (−4.47)			−0.050 *** (−5.25)	−0.331 *** (−5.39)	−0.381 *** (−6.02)
GOV	−0.065 ** (−2.26)	−0.262 *** (−3.64)			−0.094 *** (−3.41)	−0.748 *** (−4.31)	−0.841 *** (−4.71)
CON	−0.005 (−0.39)	0.037 (1.33)			−0.001 (−0.09)	0.085 (1.28)	0.084 (1.20)
FDL	−0.003 ** (−2.34)	−0.007 ** (−2.01)			−0.004 *** (−2.90)	−0.022 ** (−2.31)	−0.026 ** (−2.56)
PGDP2	−0.028 *** (−4.10)	−0.056 *** (−3.43)			−0.035 *** (−5.07)	−0.181 *** (−4.60)	−0.216 *** (−5.24)
EDU	−0.211 *** (−5.19)	−0.403 *** (−3.68)			−0.265 *** (−5.92)	−1.316 *** (−4.76)	−1.581 *** (−5.38)
rho			0.616 *** (23.72)
sigma2_e				0.001 *** (39.48)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.019	0.019	0.019	0.019	0.019	0.019	0.019
Number of id	274	274	274	274	274	274	274

**, and *** represent significance at the 5%, and 1% levels, respectively; the values represent standard errors.

, Column (7) are both statistically significant. However, the signs of these coefficients differ, suggesting that there is some transmission effect in the overall industrial structure upgrading, but a masking effect is present, with the net effect being 13.4%. This study assumes that (3) does not hold. The masking effect of industrial structure upgrading weakens the DE’s ability to suppress CEI, suggesting that while the DE reduces CEI, it also hinders overall industrial structure upgrading. The impact of the DE on overall industrial upgrading is more reflected in improving production efficiency, rather than driving the transformation of industries towards high-value-added, high-tech, and green sectors. The time lag between DE development and industrial structure transformation may be longer than our panel period allows, leading to weaker observed mediating effects.

The total effect coefficients in

Table 20. TCI spatial Durbin regression results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	TCI	TCI	TCI	TCI	Direct Effect	Indirect Effect	Total Effect
DE	−0.226 *** (−4.14)	−0.634 *** (−4.54)			−0.240 *** (−4.36)	−0.807 *** (−4.80)	−1.047 *** (−6.28)
GOV	1.227 *** (7.61)	−0.917 ** (−2.29)			1.203 *** (7.85)	−0.840 * (−1.82)	0.363 (0.80)
CON	0.564 *** (8.97)	−0.321 ** (−2.14)			0.565 *** (9.48)	−0.252 (−1.47)	0.313 * (1.82)
FDL	0.049 *** (6.22)	0.111 *** (5.66)			0.052 *** (6.77)	0.143 *** (6.24)	0.195 *** (7.88)
PGDP2	−0.244 *** (−6.40)	−0.139 (−1.54)			−0.249 *** (−6.61)	−0.220 ** (−2.06)	−0.469 *** (−4.61)
EDU	1.493 *** (6.58)	−1.330 ** (−2.20)			1.475 *** (6.69)	−1.234 * (−1.67)	0.241 (0.31)
rho			0.179 *** (5.43)
sigma2_e				0.039 *** (40.45)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.061	0.061	0.061	0.061	0.061	0.061	0.061
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

(7) and

Table 21. Spatial Durbin regression results for CEI and TCI.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	CEI	CEI	CEI	CEI	Direct Effect	Indirect Effect	Total Effect
TCI	−0.000 (−0.08)	0.049 *** (5.72)			0.005 (1.47)	0.121 *** (5.87)	0.126 *** (5.78)
DE	−0.026 *** (−2.63)	−0.108 *** (−4.26)			−0.039 *** (−4.12)	−0.305 *** (−4.94)	−0.344 *** (−5.39)
GOV	−0.062 ** (−2.11)	−0.334 *** (−4.59)			−0.097 *** (−3.50)	−0.923 *** (−5.22)	−1.021 *** (−5.61)
CON	0.016 (1.39)	0.048 * (1.77)			0.022 * (1.91)	0.143 ** (2.22)	0.165 ** (2.45)
FDL	−0.004 *** (−2.58)	−0.009 ** (−2.46)			−0.005 *** (−3.25)	−0.027 *** (−2.77)	−0.032 *** (−3.03)
PGDP2	−0.023 *** (−3.30)	−0.048 *** (−2.96)			−0.029 *** (−4.07)	−0.154 *** (−3.94)	−0.183 *** (−4.44)
EDU	−0.200 *** (−4.87)	−0.468 *** (−4.27)			−0.260 *** (−5.81)	−1.462 *** (−5.29)	−1.721 *** (−5.88)
rho			0.616 *** (23.80)
sigma2_e				0.001 *** (39.49)
Individual Fixed	YES
Time Fixed	YES
Observations	3288	3288	3288	3288	3288	3288	3288
R-squared	0.004	0.004	0.004	0.004	0.004	0.004	0.004
Number of id	274	274	274	274	274	274	274

*, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively; the values represent standard errors.

(7) are also significant, but their signs differ, indicating the presence of some mediation effect. Overall, this suggests a masking effect, with the effect ratio being 31.48%. This study assumes that (4) does not hold. In conclusion, the DE can suppress CEI through industrial structure upgrading and rationalization, but there exists a masking effect. This may be because, while the DE stimulates growth in consumption and low-end service sectors, it does not necessarily drive the development of high-value-added industries. Many traditional industries face high costs and technical bottlenecks in technological transformation, and the DE has not quickly advanced industrial upgrading to higher levels. Although the DE improves production efficiency, it also leads to short-term technological unemployment and increased social costs, which limit the advancement and upgrading of industrial structure. And certain regions—especially resource-based or underdeveloped cities—may face institutional and capacity constraints that limit the digital economy’s ability to drive seniorization of the industrial structure.

4.8. Analysis of Threshold Effects

Table 22. Threshold effect test.

Threshold Variables	Threshold Number	p-Value	Threshold Value	Boundary Value
				10%	5%	1%
CE	1	0.048	0.0326	27.6839	33.9259	43.6923

Note: Self-organized findings.

indicates that the DE exhibits a single threshold effect on CEI, which is further confirmed by the threshold effect graph in Figure 7. In

Table 23. Threshold for regression.

Variables	(1)
	CEI
GOV	−0.066 * (−1.94)
CON	0.226 *** (15.81)
FDL	−0.025 *** (−13.99)
PGDP2	−0.120 *** (−24.54)
EDU	−0.430 *** (−7.51)
DE < 0.0326	−2.028 *** (−5.70)
DE ≥ 0.0326	−0.110 *** (−8.18)
Constant	2.209 *** (24.94)
Observations	3288
Number of id	274
R-squared	0.322

* and *** represent significance at the 10% and 1% levels, respectively; values in parentheses indicate Z-values.

, the inhibitory effect on CEI peaked when the DE threshold was below 0.0326. However, when the value exceeded the threshold, the suppression effect weakened. This study assumes the validity of (5). The potential reason for this could be that the DE exhibits an optimal scale effect. Once the critical value is exceeded, traditional industries in high-carbon emission areas struggle to rapidly adapt to digitalization and low-carbon technologies, resulting in a weakened effect.

5. Discussion, Conclusions, and Recommendations

5.1. Discussion

This study provides new evidence on how the DE influences urban CEI in China through spatial spillovers, threshold effects, and mediating mechanisms. The findings indicate that the DE significantly reduces CEI not only within cities but also across neighboring regions, with spillover effects most evident within a 500 km range. Furthermore, the results reveal regional heterogeneity: while eastern and western cities benefit from strong carbon reduction effects, central regions show weaker or even reversed outcomes due to reliance on high-carbon industries. Resource-based cities also demonstrate stronger spillover effects compared to non-resource-based ones, underscoring the particular importance of digitalization in promoting industrial transformation in these areas.

These results enrich the existing literature on the DE and environmental sustainability. Previous studies have mainly emphasized efficiency gains, industrial upgrading, or green innovation as the main pathways for digitalization to reduce carbon emissions. Our findings confirm these effects but also reveal a “masking effect” of industrial structure upgrading, suggesting that digital development improves production efficiency but does not automatically drive green structural transformation. This nuance helps explain why the digital economy’s carbon reduction potential may be uneven across regions, especially in developing countries with heavy industrial dependence.

From an international perspective, differences are also noteworthy. Evidence from OECD countries and the European Union shows that digitalization not only enhances efficiency but also accelerates industrial restructuring toward green and high-tech sectors, yielding more consistent reductions in emissions. By contrast, in China, the digital economy’s effect remains more efficiency-driven, and the transition to low-carbon industrial structures is slower. These differences highlight the path-dependent nature of the digital economy’s environmental effects, shaped by industrial base, policy environment, and institutional context. Our findings thus suggest that while digitalization is a universal driver of sustainability, its actual impacts vary substantially between developed and developing economies.

5.2. Conclusions

This study adopts a comprehensive empirical strategy, including a benchmark model, SDM, two-region SDM, and robustness and endogeneity tests, as well as spatial mediation and threshold effect analyses. The research systematically examines both the direct and spatial spillover effects of the DE on CEI, delineates the attenuation boundary of spatial externalities, and assesses heterogeneity across regions and city types (resource-based vs. non-resource-based). Furthermore, it evaluates the mediating role of industrial structure advancement and upgrading. The key findings are summarized as follows:

(1) Spatial autocorrelation exists. Global and local Moran’s I and Geary’s C index tests show that CEI exhibits increasing spatial clustering over time, with H–H clusters in the eastern area and L–L clusters in the central region. (2) There is a significant carbon reduction effect. In both baseline and SDM regressions, the DE significantly reduces CEI. The effect is most pronounced in the eastern and western regions, while in the central part of China, because of the industrial structure and resource dependence, this effect is weaker and even reversed. The DE exhibits a stronger spillover effect in resource-based cities, fostering technological innovation and green development. Moreover, the spatial spillover effect reaches its peak when the distance threshold is below 500 km. (3) The threshold effect exists. The DE’s effect on CEI is optimal at a threshold value of 0.0326, beyond which the effect weakens. (4) The masking effect exists. The mediating role of industrial upgrading and advancement reveals a masking effect. In the short term, the DE primarily enhances production efficiency without significantly accelerating the transition to green, high-tech sectors. This limits its impact on structural transformation and suppresses the full realization of carbon reduction potential. (5) Although this study focuses on China, the analytical framework combining spatial econometrics, threshold regression, and mediation modeling can be applied to other emerging economies undergoing rapid digitalization. The findings thus provide valuable insights for countries facing similar challenges of balancing digital growth with low-carbon transformation.

5.3. Recommendations

Based on the above findings, several targeted policy recommendations can be proposed:

(1)

Promote regionally differentiated digital development strategies

The results show that eastern and western cities benefit more strongly from the carbon reduction effects of the DE, while central cities experience weaker or even reversed outcomes. Therefore, policymakers should adopt differentiated strategies: eastern and western regions should consolidate digital infrastructure and accelerate digital–green synergies, while central regions should be supported with industrial restructuring funds, clean energy incentives, and capacity-building programs to ensure that digitalization contributes to low-carbon development rather than reinforcing carbon-intensive structures;

(2)

Leverage resource-based cities as demonstration zones

Resource-based cities show stronger spillover effects, indicating that digital transformation in these areas can radiate benefits to surrounding regions. Policymakers should prioritize such cities as pilot zones for digital-driven green transformation by supporting renewable energy deployment, digital monitoring of industrial emissions, and the development of smart mining and eco-restoration platforms;

(3)

Avoid overconcentration and diminishing returns of digital investment

The threshold effect indicates that beyond a certain level, the marginal benefits of digital economy development diminish. To avoid inefficiency and regional imbalance, resources should be reallocated to underdeveloped cities to expand the inclusiveness of digital infrastructure, ensuring that all regions can benefit from digitalization’s carbon reduction potential;

(4)

Integrate digitalization with industrial upgrading and green finance

Since industrial structure upgrading currently shows a “masking effect,” policymakers should ensure that digital development is coupled with structural transformation policies. This includes providing fiscal and tax incentives for high-tech and green industries, strengthening environmental regulations for carbon-intensive sectors, and promoting green finance instruments that direct capital into low-carbon projects;

(5)

Encourage international cooperation and knowledge sharing

Given the path-dependent nature of the environmental impacts of the digital economy, China should actively participate in international cooperation, drawing lessons from OECD countries, the European Union, and other advanced economies, while at the same time sharing its own digital low-carbon practices with other emerging economies. Such efforts would contribute to building a global collaborative framework in which digitalization serves as a key driver for sustainable development and the achievement of carbon neutrality goals.

6. Limitations and Future Research Directions

Despite providing valuable insights into the relationship between the DE and CEI, this study has several limitations that offer opportunities for further research.

First, the measurement of core variables—particularly the DE index and CEI—relies on city-level statistical data, some of which are interpolated due to data gaps. This may affect the accuracy of estimation. Future studies could leverage high-frequency or micro-level data, such as firm-level surveys, remote sensing data (e.g., nighttime light intensity), or real-time digital transaction records, to improve measurement precision.

Second, although endogeneity is addressed using the 2SLS approach, the instrumental variable—based on geographic proximity to Hangzhou—may not fully capture exogenous variation. Future research could explore natural experiments or difference-in-differences designs to strengthen causal inference.

Third, the spatial weight matrix used in the SDM is constructed based on economic–geographic distance. While this approach is reasonable, it may not fully reflect actual inter-city economic interactions. Future work could explore alternative matrices—such as trade flows, supply chain linkages, or digital infrastructure networks—to better capture spatial dependence.

Fourth, future research could extend this analysis to cross-country comparative studies, thereby deepening the understanding of how digital transformation interacts with diverse institutional and developmental contexts in shaping global carbon reduction outcomes.