The digital economy is vigorously emerging worldwide. US academic Tapscott first conceptualized the digital economy at the beginning of internet development in the 20th century (Tapscott, 1996). Since then, scholars have increasingly directed their attention toward the digital economy. Studies have been conducted to define and measure DIGDE from different perspectives. According to Bukht et al. (2017), DIGDE is a type of economic production that is derived entirely from or that mostly relies upon digital technology, where digital goods or services are the base point.

Scholars in China and elsewhere have made many useful attempts to measure DIGDE. These attempts are typically divided into two groups. The first consists of direct methods, which estimate the corresponding DIGDE index to examine and compare the DIGDE index within each region (Eurostat, 2017; ITU, 2022; UNCTAD, 2021). The second category consists of construction methods, in which a multidimensional evaluation index system is constructed based on different perspectives (Cheng et al., 2023; Lin and Huang, 2023) and is subsequently used to measure DIGDE by assigning weights to the indicators.

Research on CE has focused on carbon accounting along with differences in the spatial distribution of CE and influencing factors. There have been various concepts and methods of carbon accounting. For example, in 2011, the Chinese Academy of Sciences (CAS) started the “Climate Change: Carbon Budget and Relevant Issues” project to build a visualization system by integrating various utilization sector data to obtain the consumption factors of different energy types. The China Emissions Accounting and Datasets (CEADs) team used these CE factors to calculate and publish the corresponding CE inventory. Cai et al. developed a bottom-up urban greenhouse gas (GHG) accounting approach that can systematically reduce the uncertainty in emission variables and activity levels (Cai et al., 2018; Liu et al., 2021).

Numerous studies have confirmed that the distribution of CE varies significantly in space (Tong, 2020; Pan et al., 2023; Xu et al., 2023) and is determined by factors such as government intervention (Xiang et al., 2023), energy intensity (Chai et al., 2023), renewable energy (Azam et al., 2022), the economic output trend (Song et al., 2022) and industrial development (Cai et al., 2023). The research methods are focused on quantitative analysis. For example, Wang et al. (2023) employed structural decomposition analysis (SDA) as a method to assess the contributing factors affecting bilateral CE in 30 Chinese provinces, and they found that the technology effect can suppress bilateral CE, while the demand effect promotes bilateral CE. Azam et al. (2023a) used a panel autoregressive distributed lag (ARDL) found that negative synergy is perceived between CE and agricultural productivity.

(ARDL) model and found a negative synergy between CO₂ emissions and agricultural productivity.

Established theoretical studies have argued that the carbon reduction impact of DIGDE is formed based on several aspects with the addition of digital technologies. At the governance level, digital governance theory holds that remote sensing technologies, big data, and cloud computing applications can increase the precision and efficacy of governmental environmental control (Yang et al., 2021) to enhance ecological governance (Thierer and Castillo, 2015), contributing to the realization of CE reductions. The flow of information between policymakers and the masses has been changed by enabling information interoperability and sharing between the government and society through digital media as a result of digital technology (Nulman and Ozkula, 2016; Bai et al., 2023). Simultaneously, the distribution of interactive information and internet environmental monitoring enable innovative interactive contact mechanisms between society and the government, promoting collaborative governance among all parties in the preservation of the ecosystem (Yang et al., 2020), which will enhance the efficiency of government governance (Chen et al., 2023b), and jointly promote the development of a low-carbon economy. From the perspective of energy efficiency, DIGDE can overcome limitations of time and space, accelerate the flow of factors and reduce energy consumption during transmission (Zhang et al., 2022). Meanwhile, digital technology strengthens green finance, accelerates the adoption of renewable energy, further promotes energy transformation (Han and Li, 2022) and improves energy use efficiency, which curbs CE. The findings of some research support this view. For example, Xie et al. (2024) found that DIGDE increases CE in the short term and exerts a carbon reduction effect in the long term. Wang et al. (2022) found that DIGDE is beneficial for reducing urban CE. Cheng et al. (2023) found that DIGDE reduces carbon emission intensity when the DIGDE index exceeds 0.419. Ma et al. (2022) conducted a study at the provincial level and found that DIGDE in China reduces the level of CE, while investments in research and development related to digitization also have a dampening effect on CE. Niu et al. (2024) found that DIGDE affects the transfer of CE between regions and reshapes resource trade relations.

Many beneficial explorations of DIGDE and CE have been conducted in the available literature, laying a rich foundation for our study. This study is based on the typical fact that DIGDE affects CE, and cuts in from the spatial perspective. Compared with the existing studies, this study focuses on SSE, by combining SDM with the mediating effect model to develop the study of the transmission mechanism from the spatial perspective. It also innovatively constructs a spatial weight matrix to empirically examine SSE of DIGDE on CE under different factors, such as human capital, service industry development and information development. To provide empirical support and policy references to give full play to the advantages of DIGDE, maximize support for the realization of spillover effects, and further develop its positive role in promoting synergistic low-carbon development in the region.

DIGDE has led to a series of technological innovations (TEI) and management innovations that have been collected, integrated and distributed through the internet to maximize the effective use of resources. DIGDE has an impact on CE at three main levels. The first is the emission reduction effect of optimal resource allocation. DIGDE integrates information on production factors and resources through digital technology, optimizing resource allocation and the energy use structure, and thus reducing CE. At the same time, by promoting multiparty cooperation and group agglomeration, DIGDE has a positive externality effect on neighboring regions and even the whole economic system, i.e., it has an SSE. The second is the emission reduction efficacy in the low-carbon development model. DIGDE has broken the spatial and temporal barriers to production activities and has facilitated the regional circulation of production factors (Li and Wang, 2022), bringing positive externalities to overall output through local innovation activities (Park, 1995). The “digitalization of environmental sustainability” is promoted using monitoring technologies, enabling the generation of real-time CE data (Kloppenburg et al., 2022). Third, there is the emission reduction effect of environmental governance model innovation. Digital technologies are widely used in environmental governance, weakening geographic and organizational boundaries through digital platforms, attracting various stakeholders to construct and solve problems (Ozman and Gossart, 2017), and promoting the formation of informal environmental regulation dominated by the networked public (Certoma, 2022). Under the new pattern of information opening and sharing, through the role of competition and demonstration effects, the positive SSE of DIGDE on regional high-quality development can be brought into play, and the green transformation ability of surrounding regions can be improved. Thus, the following hypotheses is proposed.

H1

There is SSE on the impact of DIGDE on CE.

DIGDE promotes changes in research and development (R&D) and innovation paradigms, provides new means and channels for innovation information access, and enhances innovation efficiency and quality (Lai et al., 2022). TEI on the energy supply side can accelerate the development of clean energy and the use of low-carbon technologies, promote the formation of new industrial and value chains, and facilitate the sharing and dissemination of low-carbon technologies and experiences. Thus, there are so-called innovation spillover benefits. TEI on the energy consumption side can improve energy efficiency. The application of low-carbon technologies in the transportation, construction, and chemical industries, directly contributes to CE reduction and drives the upgrading of green technologies in neighboring regions through cross-regional environmental collaboration.

At present, the resource endowment and scientific and technological development are hindering regional economic development and the realization of CE reduction targets in China. The realization of the goal of ecological civilization construction not only requires the guidance of TEI but also places higher demands on the energy supply system. DIGDE brings digital support to energy consumption structure (ECS) adjustment, which can support the government in quickly perceiving and making quick decisions by creating new tools to guide the real-time flow of energy factors (Ferreira et al., 2023). The widespread use of clean energy and new technologies can enhance the use of renewable energy and reduce the total CE from economic activities, ultimately forming a diversified and low-carbon energy supply pattern. Since the industrial chains and energy supply chains of neighboring regions are interrelated, the industrial adjustment and transfer brought by ECS optimization in a region can affect CE of neighboring regions through the cross-border flow of energy and production factors. Through the comprehensive analysis conducted above, the following hypothesis is proposed.

H2

DIGDE can influence CE by driving TEI and optimizing the ECS.

The stochastic impacts by regression on population, affluence, and technology (STIRPAT) model is a common model for analyzing the influence of economic elements on the environment. This model can take into account the unequally proportional influence of human factors on the environment and has good scalability. This study extends the STIRPAT model and constructs a baseline regression model by incorporating the theoretical analysis above as follows: ln ⁡ C E i t = α 0 + β 1 l n D I G D E i t + β 2 l n F D i t + β 3 l n L Y i t + β 4 l n U R i t + β 5 l n D I G D E i t + β 6 l n D I G D E i t + μ i + σ t + ε i t (1)

Here, in Eq. 1 lnCE _it and lnDIGDE _it denote the level of CE and DIGDE at year t in region i, respectively. lnGOV, lnOS, lnFD, lnUR and lnLY denote government financial support, the degree of marketization, the degree of openness to the outside world, the urbanization level and the level of per capita income, respectively. The area fixed effect is denoted by μ, the time fixed effect is denoted by σ, and the random disturbance term is denoted by ε.

Not only are CE directly influenced by the policies and economy of a region but they are also influenced by related factors in surrounding areas (Chen et al., 2023a). To adequately consider these influencing factors, this work studies the influence of DIGDE on CE by establishing an SDM. Moran’s I can test whether there is spatial autocorrelation in the data, and SDM can be used to explore its specific correlation. The specific construction of SDM requires a series of tests, the use of Hausman’s test can determine whether the model should be selected as a random-effects model (REM) or a fixed-effects model (FEM) (Azam et al., 2023b), the LR test can be used to further determine the fixation of the individual or time or both, and the combination of the WALD test can ultimately select the most appropriate model. ln ⁡ C E i t = α 0 + ρ ∑ j = 1 n W i j l n C E i t + β 1 l n D E C i t + ∑ j = 1 n W i j l n D E C i t γ 1 + μ i + σ t + ε i t (2)

In Eq. 2 lnDEC is the explanatory variable, which includes DIGDE and the corresponding control variables, ρ ₀ denotes the spatially lagged regression coefficient. The remaining parameters are set as in Eq. 1.

Based on the previous theoretical analysis, referring to the results of existing research on the impact factors of DIGDE on CE (Grigorescu et al., 2021; Williams, 2021; Jiang et al., 2023; Pan et al., 2023; Wang et al., 2023), this study considers four major factors: geographic proximity, human capital development, service industry development, and information development. Four spatial weight matrices are used to process the SDM in this study. The first is the adjacency spatial weight matrix (W ₁ ), which uses 0 and 1 to mark the spatial adjacency between regions. W _ij = 0 when region i, and region j are neighboring and W _ij = 1 when region i and region j are not neighboring.

The second is the human capital matrix (W ₂ ), which takes the average annual employment in each province as a measure. When the level of human capital in two regions is similar, the greater weight of the two regions is considered, assuming that the levels of human capital in region i and region j are h _i vs. h _j .

The third is the service industry development matrix (W ₃ ), which, as a measure, is calculated by using the proportion of output value of the tertiary industry to GDP as the service industry development level in the two regions, assuming that the level of development of the service industry in the two regions is s _i and s _j .

The fourth is the information development level matrix (W ₄ ), which, as a measure, is calculated by using the average annual telecommunication business revenue in each province, assuming that the level of information development in the two regions is d _i and d _j .

The matrix construction formula is as follows. W i j = 1 / z i − z j (3)

In Eq. 3 z _i denotes different variables in spatial weight matrix setting h _i , s _i , and d _ij ; z _j denotes h _j , s _j and d _j .

To verify whether TEI and the ECS act as mediating variables in accordance with the previous theoretical hypotheses, this study conducts a multiple mediating effect test based on the stepwise regression method combined with the SDM.

First, the regression coefficients of CE and DIGDE are tested to verify whether DIGDE has a direct impact on CE, as in Eq. 1.

Second, whether there are direct effects of DIGDE on the mediating variables is examined in Eq. 4. ln ⁡ C E i t = α 0 + ρ ∑ j = 1 n W i j l n C E i t + β 1 l n D E C i t + ∑ j = 1 n W i j l n D E C i t γ 1 + μ i + σ t + ε i t (4)

Finally, the indirect and total effects of DIGDE and the mediating variables on CE are examined. ln ⁡ M i t = α 0 + ρ ∑ j = 1 n W i j l n C E i t + β 1 l n D E C i t + ∑ j = 1 n W i j l n D E C i t γ 1 + μ i + σ t + ε i t (5)

Where M _it denotes the mediating variable in Eq. 5.

Data are primarily taken from the China Statistical Yearbook, the National Bureau of Statistics, and relevant regional statistical yearbooks. The Peking University Digital Inclusive Finance Index comes from the Institute of Digital Finance, Peking University. Due to excessive missing data for Tibet, Taiwan, Macao and Hong Kong, they are not discussed in this study.

Missing data are filled in through linear interpolation. To avoid pseudoregression and eliminate heteroskedasticity, all variables are transformed using logarithmic scales.

Table 2 lists the descriptive statistics of each variable made in this paper, Table 3 lists the descriptive statistics of each variable: the maximum values of lnDIGDE and lnCE are 6.2405 and 11.7487, respectively, and the minimum values are 2.2721 and 8.1360, which are basically consistent with the measurements observed in the literature. lnOF has the maximum value of 7.5183 and the minimum value of 0.4164, which indicates that the level of openness to the outside world is uneven among different regions, and polarization is serious. This may be due to different geographic locations; coastal areas have developed freight transportation, so their level of openness to the outside world is higher. The standard deviation of lnLY is the largest at 1.3058, and the standard deviation of lnOS is the smallest at 0.2620, which indicates that per capita income levels vary widely across different regions in the sample, and the degree of marketization does not vary greatly.

This study explores the agglomeration features of DIGDE and CE from 2011 to 2020 based on Moran’s I (Table 4), the findings of which demonstrate that Moran’s I is positively significant under W ₁ . It is inferred that the close connections among different cities and their generated correlations can affect the spatial correlation of DIGDE and CE.

The baseline regression model of Eq. 1 is first estimated, and the benchmark test results are shown as the regression findings in Table 5. The three columns show the results of mixed OLS, fixed effect and cluster standard errors. By comparing the model results, it can be seen that the model’s goodness of fit, significant levels of variables and coefficients do not change much in the three regressions, which indicates that the variables entered into the model are relatively stable. That is, the inhibitory effect of DIGDE on CE is stable and reliable. In summary, the regression results confirm Hypothesis 1 that DIGDE can suppress CE. Further spatial studies can be carried out on this phenomenon.

To determine the most effective model for exploring the relationship between DIGDE and CE, this study uses the LM, Wald, and LR tests (Chen et al., 20233), indicating that the SDM with time fixed effects is the best choice, the Hausman test supports the fixed effect model (Zhao and Wang, 2022) (Table 6). The partial differential method is applied to decompose the impact of the local region and neighboring regions into direct effects, indirect effects and total effects (LeSage and Pace, 2009) (Table 7).

The coefficients of the impact of DIGDE on CE and the SSE are both significantly negative under W ₁ . Compared with Table 6, the impact of DIGDE on CE will be underestimated when the SSE is ignored, which is not conducive to effective regional environmental regulatory policies and DIGDE. The coefficient of the indirect effect of DEIGDE on CE is −1.4450, which is 1.7397 times the direct effect. This result indicates that DIGDE in a region can influence CE in neighboring regions, and when the level of DIGDE in a region increases by 1%, its inhibitory effect on CE in neighboring regions is 1.7297 times higher than that in the region. This phenomenon is due to the accelerated development of internet trading platforms, which accelerates the cross-regional flow of production factors. Through open sharing, the development of internet trading platforms promotes the productivity of neighboring regions, thus contributing to the optimization of resource allocation. It can be assumed that if neighboring regions also accelerate DIGDE, the overall level of CE reduction in an area will be improved, thus forming a virtuous circle of the “snowball effect” between regions.

Under W ₁ the coefficient of the control variable lnLY exhibits a positive effect on CE in surrounding areas and a negative effect on CE in region. The scale and development of production activities in an area are largely determined by the level of consumption of residents. High-income groups tend to have a higher demand for environmental standards and affordability. This higher demand has prompted enterprises to implement green production and sustainable development strategies and shift industries that are not environmentally friendly to neighboring areas. lnUR increases CE in a region, while it exhibits a lowering effect on adjacent areas. The rise in the urbanization level implies a more pronounced population agglomeration effect, which is accompanied by a shift in the labor force and a change in economic activity patterns, thus increasing transportation CE due to population migration. The lnOS and lnGOV variables increase local CE, and because the environmental effects have a relatively long and low return cycle, local governments tend to be more likely to invest in fields with faster economic return. If the direction of marketization and government financial support is oriented toward increasing productivity, then it will promote production scale expansion, which is not conducive to ECS and CE reduction (Shao et al., 2013). Furthermore, the promotion of new environmental protection technology industries and the impact of government regulation on the environment often have a certain time lag. Therefore, current period market-based reforms do not necessarily have a significant dampening effect on CE. However, sharing infrastructure with neighboring regions can effectively avoid duplication of investment in construction and waste of resources and reduce the land occupation, energy consumption and material extraction required for new projects, thus helping neighboring cities save energy and reduce CE.

The results of the stepwise regression and the decomposition of the mediating effects are listed in Tables 8, 9, and the results in model (2) satisfy the prerequisites for the subsequent stepwise regression in the theory of Judd and Kenny, (1981).

Regarding the mediating transmission mechanism of TEI, DIGDE can significantly promote TEI, and after adding the mediating variable TEI, CE are shown to be significantly slowed by DIGDE and have a significant SSE. This result indicates that TEI holds as a mediating variable. Regarding the mediating transmission mechanism of the ECS, in model (3), the coefficient of the effect of DIGDE on the ECS is significantly negative. DIGDE on CE is negative, and the ECS on CE is positive in model (4). These results indicate that DIGDE can promote CE reduction, and at the same time, the increase in the proportion of coal consumption leads to higher CE levels, which once reaffirms that China’s “high-carbon” ECS with an abnormally high reliance on coal is an important reason for the hindrance in CE reduction (Shao et al., 2019). The coefficient shows that DIGDE has a strong promoting effect on TEI, and when the intermediary variable TEI exists, DIGDE has a more positive inhibiting effect on CE. DIGDE has driven the intelligent transformation of traditional industries and improved the efficiency of resource utilization, reducing the use of fossil fuels in the production process, and achieving the effect of saving energy to reduce emissions at the source. The SSE of the mediation effects decomposition results in Table 8 is also consistent with the expected assumptions. That is, DIGDE can lead regional green collaborative development through TEI and by optimizing the ECS.

The CE estimation method above uses regional CE for measurement, and there are still many studies in the literature that use per capita CE to measure CE levels. In Table 10, when lnPCE is used as the dependent variable, the mediating effect and SSE are still valid, which confirms the robustness of the results above.

In econometric regression, to obtain consistency in the effects, it is important to address possible endogeneity. Different from traditional ordinary least squares (OLS) regression, the SDM makes it possible to obtain estimates that are not biased by amplification, thus avoiding endogeneity due to omitted variables (LeSage and Pace, 2009). Drawing on Wang and Guo (2023), this study uses the generalized spatial two-stage least squares (GS2SLS) model to control for the endogeneity problem of the key variables and lags the explanatory variables by one period. The regression results in Table 11 show that after mitigating the potential endogeneity problem, the study’s conclusions still hold, and the mediating effect remains.

In Table 12, the SSE of DIGDE on CE also exists among regions with similar human capital, service development, and information development and is similar to the effects under W ₁ . Additionally, the indirect effects are all larger than the direct effects, confirming H3. The coefficient of the indirect effect of DIGDE on CE is the largest under W ₄ .

In the context of digitalization and intelligence, human capital has become an important resource for regions, and innovative and high-tech companies have improved their innovation efficiency and green transformation capabilities by adjusting workforce involvement through effective talent management (Zahoor et al., 2022). Green enterprises promote carbon reduction in regions with higher environmental standards and more sophisticated energy-saving technologies, and can play a positive role in leading and regulating the development of regions with similar human capital. Meanwhile, DIGDE involves the Internet of Things and other fields with high intensity R&D investment, which can produce strong SSEs. These effects change the development model of the service industry through the intellectual capital and human capital needed by the service industry, and they promote its digital transformation. Through online platforms and applications, many traditional processes can be optimized, reducing energy consumption and CE in physical service processes. Regions with similar levels of service industry development have similar industrial structures and usually have closer flows of technological elements and industrial interconnection, which can promote green synergy through technology promotion and resource sharing. DIGDE drives the efficient operation of material flow and technology flow with information flow, and it promotes the optimal allocation of resource elements between industries. The symbiotic union of different types of industries is becoming increasingly common, and industrial integration is deepening, prompting the gradual adjustment of the industrial structure to be high grade, low carbon and green. Therefore, in regions with similar human capital and service development, DIGDE can exert an SSE that reduces CE. In addition, regions with similar levels of information technology development have high levels of networked synergy. Open platforms based on information technology provide a borderless space for information sharing, which fully reduces information asymmetry in the process of rapid information flow (Asongu et al., 2017), making the transmission of green technology ideas and environmental protection more effective. Therefore, compared with W ₂ and W ₃ , it is easier to bring into play the SSE of DIGDE on CE under the W ₄ matrix.

Given the typical fact that DIGDE affects CE, this study starts from the spatial perspective, and based on the panel data of 30 provinces and regions in China from 2011 to 2020, it innovatively constructs different spatial weight matrices on the basis of measuring the level of DIGDE and combines SDM with the mediation effect model to investigate the mechanism of the impact of DIGDE on CE and the multiple SSE from the spatial perspective. The following conclusions were drawn from this study. First, there is obvious spatial heterogeneity in DIGDE. The eastern region has the highest overall average level of DIGDE, which has reached a certain scale and height. The rapid growth of DIGDE in the central region, with the highest average annual growth rate, showing great potential and opportunities for development. The average annual growth rate of the western region is slightly lower than that of the central region, it also shows a stable development trend and broad development space. Second, DIGDE can suppress CE and have a significant SSE. This means that with DIGDE, not only can CE be reduced directly, but its influence can also be transferred between regions and have a dampening effect on CE in neighboring regions. DIGDE can indirectly reduce CE by driving TEI and optimizing ECS. Third, the impact of DIGDE on CE is also influenced by other factors, under the role of human capital, service industry development and the information development matrix, DIGDE has a negative SSE on CE. Regions with a similar level of information development are more likely to exert SSE of DIGDE on CE. This further emphasizes the important impact of synergies between DIGDE and other elements of development on CE.

Mitigating climate change and reducing CE is the common responsibility of all mankind. Based on the findings, the policy implications of this study are as follows.

The policy insights obtained from this paper are as follows. First, DIGDE is conducive to reducing CE, and in addressing the challenges of climate change, we should continue to increase the level of DIGDE. Continuously unleashing the dynamism of demand, including investment and information consumption, in DIGDE, and give full play to the leading role of DIGDE in green development. In response to SSE, interregional economic ties should be strengthened, and efforts should be made to narrow the gap in DIGDE between regions by exploiting different regions and forming a pattern of coordinated regional development by building a mechanism of synergistic development and complementary advantages between regions. Second, we should fully consider the transmission mechanism through which DIGDE affects CE and focus on improving the level of TEI and strive to drive the rise of low-carbon industries through the development of digital technology. The restructuring of ECS should be accelerated, the level of conversion and application of new energy should be improved, and a clean, low-carbon, safe and efficient energy system should be built. The government can strengthen the awareness of energy saving and emission reduction of enterprises by strengthening supervision and forcing them to optimize their own ECS. Third, considering the strongest SSE of DIGDE on CE under the information development level matrix, in exploring the practice of DIGDE and CE reduction, we should focus on learning from regions with similar levels of information development to quickly and effectively accumulate experience and give full play to the important role of information development in driving regional synergistic development.

The limitations of this study are as follows. First, digital technology is constantly iteratively developing, and methods of evaluating DIGDE can be further explored. Second, this study focuses on the impact of TEI and the ECS. Subsequent studies can also proceed from the perspectives of environmental regulation and economic ag-glomeration to further explore the impact path of DIGDE on CE. Finally, from the perspectives of service industry development, human capital and information development, this study examines the SSE of DIGDE on CE. Follow-up studies can also be carried out from the perspectives of other context elements.