Level of Integration of Real and Digital Economies: Effects and Mechanisms of Environmental Pollution Impacts

Abstract

As global economic development advances, the constraints of traditional growth paradigms, particularly the escalating challenge of environmental pollution, have become increasingly evident. In this context, the deep integration of the digital and real economies (IDE) has emerged as a promising approach to sustain economic expansion while addressing environmental concerns. Drawing on panel data from 30 Chinese provinces throughout 2008–2022, this study employs the entropy weight method and the coupling coordination degree model to quantify the levels of IDE and pollution. A two-way fixed-effects regression framework is then applied to assess the relationship between IDE and environmental pollution and to uncover potential mediating mechanisms. The principal findings are as follows: (1) The integration level of the digital and real economies has a suppressive effect on environmental pollution, with this effect exhibiting significant regional heterogeneity. (2) The deep IDE facilitates the optimization of the industrial structure (IS) and the reduction in energy consumption through two intermediary channels, leading to a marked improvement in environmental quality. (3) The industrial structure exhibits a threshold effect within the mechanism, with its influence on pollution levels displaying a nonlinear model characteristic of increasing marginal effect. These results enrich the interdisciplinary nexus of environmental studies and the digital economy, offering a scientific basis for policymaking and contributing to China’s dual-carbon objectives and the global sustainability transition. Future research may explore the differentiated impacts of digital convergence under diverse policy regimes and identify strategies to maximize its environmental benefits.

1. Introduction

Over the past few decades, China has undergone rapid economic expansion [1] and has substantially advanced industrialization and urbanization. However, the traditional development model has revealed multiple shortcomings, most notably the intensification of environmental pollution [2], which has become a critical barrier to the sustainable growth of the economy of China and those of many other countries [3]. Environmental challenges, such as air, water, and soil contamination, have grown increasingly severe, resulting in significant economic losses [4] and exerting profound adverse effects on ecosystem integrity.

In this context, the digital economy has rapidly emerged in recent years [5] and has had a series of positive effects on economic development [6]. The real economy encompasses all economic activities related to production, distribution, and sales [7], while the digital economy includes all economic activities that rely on digital information, electronic transactions, and the use of digital devices [8]. The deep integration of the digital economy with the real economy has become an emerging pathway for driving green transformation. This transformation is reflected in various aspects such as production and distribution, operational efficiency, innovation models, and digital innovation, offering new opportunities to improve carbon emission efficiency and the environment [9]. Research on the coupling and coordinated growth between the real economy and the digital economy is crucial for achieving high-quality green development [10]. However, most existing studies focus on how the digital economy can facilitate the transformation and upgrading of the traditional economy [11], primarily examining the direct impact of digital-real integration on economic efficiency. The exploration of its environmental effects remains insufficient in the following areas: (1) limited research on the mechanisms through which integration affects environmental pollution via intermediary channels; (2) a lack of analysis on regional heterogeneity, particularly whether differences in regional economic foundations and digital infrastructure levels contribute to spatial environmental effects; (3) the need to verify whether the influence mechanism of digital–real integration on environmental pollution is linear or nonlinear.

This study utilizes panel data from 30 provincial-level administrative regions in China from 2008 to 2022 and employs the two-way fixed-effects model, the mediated-effects model, and the threshold model to systematically explore the environmental governance mechanisms of digital–real integration. Theoretically, this research addresses a gap in understanding the mechanisms underlying the environmental effects of digital–real integration. Practically, the findings provide a scientific foundation for the development of differentiated regional policies, contributing to the global sustainable development agenda.

2. Literature Review and Research Hypothesis

Is economic growth a solution to environmental problems, or is it the cause? Grossman and Krueger (1991) proposed the Environmental Kuznets Curve (EKC) [12], which suggests an inverted U-shaped relationship between economic growth and environmental quality, indicating that as the economy expands, environmental pollution eventually decreases. With the escalating global environmental pollution issues, the relatively optimistic EKC theory has inspired numerous studies exploring the impact of economic development on the environment, including those driven by population growth [13] and improper waste disposal [14]. In addition to traditional approaches, some scholars have begun to explore emerging trends in the environmental analysis of pollutants [15] and investigate the role of new materials in pollution treatment [16]. Furthermore, some researchers are examining the integration of the digital economy with the real economy to foster new competitive advantages in digital transformation [17]. Recent research has highlighted the potential of the integration of the digital economy and the real economy (IDE) in promoting green transformation through digital technologies such as artificial intelligence [18] and the Internet of Things [19], with digital literacy potentially accelerating the EKC inflection point through the “technology effect”.

The existing literature has primarily linked digital resources to economic growth [20], with relatively few studies exploring the connection between their integration with physical entities and environmental pollution. While previous research has recognized the economic benefits of IDE and technological advancements, such as Zhao (2014), who examined the relationship between e-government development and the digital economy and its long-term impacts on public management [21], and Xu et al. (2023), who quantitatively assessed the innovation-driven effects of China’s digital economy on the real economy at national and regional levels through econometric regression [22], their environmental impact is often regarded as a secondary or indirect effect. The studies mentioned above focus on IDE contributions to public management and innovation but overlook its potential for reducing pollution. Kannan Govindan (2023) investigated how digitalization can transform the traditional circular economy into a smart circular economy to achieve the SDGs [23], but analytical evidence on its role in mitigating environmental pollution remains limited. Rocca (2020) integrated virtual reality and digital twins into circular economy practices to address the changing demands of the industrial sector under the impact of climate change [24] but did not isolate or analyze the role of IDE in environmental outcomes. Charfeddine and Umlai (2023) conducted a meta-analysis of 166 studies examining the environmental impacts of digitization from 2000 to 2022. Most of these studies relied on outdated and oversimplified metrics to represent digitization [25]. Additionally, many studies assumed a linear relationship between IDE and environmental outcomes, neglecting threshold effects and spatial variations. For example, Mirza and Kanwal (2017) found dynamic causality between energy consumption, economic growth, and carbon dioxide emissions in Pakistan [26] but did not validate its linear or nonlinear nature. Cheng and Hu (2023) favor a nonlinear relationship between economic development and carbon emissions [27]. Such discussions should be extended to broader contexts. While the EKC posits that economic growth initially exacerbates pollution, IDE-driven technological spillovers (e.g., AI-driven efficiency) may flatten this curve by enabling ‘leapfrogging’ effects, where regions bypass traditional high-pollution industrialization stages.

This study aims to address this research gap by empirically analyzing the extent to which the latest IDE directly affects environmental pollution.

Based on this, this study proposes Hypothesis 1 (H1):

H1: 

The integration of the digital economy with the real economy contributes to reducing environmental pollution levels, and its impact may not necessarily be linear.

Most studies measuring the level of IDE integration, such as Xin et al. (2024), have assessed the level of integration between the digital economy and industries, as well as its impact on energy conservation and emission reduction, using methods like entropy-weighted TOPSIS [28]. These studies primarily focus on efficiency, with insufficient research on environmental governance mechanisms. This narrow focus highlights a key gap: the lack of systematic evidence on how IDE are structured to achieve environmental efficiency through specific mechanisms. Few studies define IDE as niche innovations that disrupt socio-technical institutions. This articulates how digital technologies are reshaping the emissions pathways of traditional industries.

In more developed economic structures, high-tech industries, such as information technology and biotechnology, tend to dominate. Compared to less developed economic structures, these industries generally exhibit higher resource utilization efficiency and lower pollution emissions [29]. IDE integrates technological and data elements into the production and operation processes of traditional industries, thereby strengthening the real economy and shaping the layout of the modern industrial system [30]. Previous studies, such as Yang et al. (2022), explored the impact of IDR on carbon emission efficiency using panel data from 274 cities in China and identified the mediating role of industrial structure (IS) upgrading [31]. Similarly, Yang and Liu (2024) viewed industrial upgrading as a mediator of green efficiency [32] but failed to quantify the broader environmental contributions resulting from IDE expansion within the context of environmental protection. By integrating entropy-weighting methods and mediation modeling, this study explicitly measures how IDE-driven structural shifts (e.g., growth of the tertiary sector) translate into environmental benefits, a mechanism that has been underexplored in previous studies. Based on this, Hypothesis 2 (H2) is proposed:

H2: 

The development of digital–real integration improves IS and reduces environmental pollution levels.

From the demand side, the development of digital–real integration fosters new modes such as online shopping, telecommuting, and online education. The networking and dematerialization of economic activities reduce energy demand and increase the reuse rate of goods [33]. Energy consumption, which has a significant impact on the environment, is addressed in studies like Suki et al. (2022), who examined energy consumption and environmental changes in Malaysia from 1961 to 2016, showing that the increased use of renewable energy slows environmental degradation [34]. Although prior studies have recognized the potential of digital technologies to improve energy efficiency [35], key limitations remain. Some studies focus on a single channel, such as Firnkorn (2015), who emphasized the role of sharing economy platforms in reducing private car use [36] but neglected their systemic impact on the energy supply chain. In terms of dynamic perspectives, for instance, Liang et al. (2023) linked natural resources and economic development to investigate whether digital transformation could alleviate resource constraints [37] but focused less on how IDE dynamically alters the energy mix over time. To fill these research gaps, this study conducts an empirical analysis with panel data to examine the mediating effect of energy consumption between IDE and environmental pollution levels, moving beyond static correlations. Based on this, Hypothesis 3 (H3) is proposed:

H3: 

The development of digital–real integration reduces environmental pollution levels by decreasing energy consumption.

Many cutting-edge studies, such as Lu et al. (2023), have shown that digital technology not only provides new opportunities for the transformation and upgrading of the real economy but also has significant positive impacts on the environment [38]. However, most of these studies do not provide an in-depth analysis of regional heterogeneity. In sustainability transition research, many scholars, such as Barbara Breitschopf (2023), emphasize technology–institutional interaction but lack a quantitative analysis of regional heterogeneity [39]. These research gaps have not prompted policymakers to tailor strategies to specific contexts, particularly in developing economies with uneven digital infrastructure.

The level of real economy development, along with factors such as regional economic level, urbanization [40], and labor force structure [41], plays a critical role in digital economy development across regions. While spatial heterogeneity in digital development has been acknowledged, the quantification of its environmental consequences remains limited.

Regional differences in various factors lead to clear spatial heterogeneity in digital economic development in China. Geographically, the improvement in environmental pollution levels should vary across regions and cannot be simplified into a linear relationship. Most studies also fail to recognize the application of innovation diffusion theory: early adopters (eastern China) face diminishing returns from incremental innovation, while late adopters (western China) may benefit from disruptive system-wide digital transformation.

By comparing the eastern, central, and western regions of China, the policy-relevant issue of how the regional digital divide affects environmental governance efficiency is explored, and feasible policy strategies are provided to support sustainability transitions. Based on this, Hypothesis 4 (H4) is proposed:

H4: 

The effect of the level of digital integration on environmental pollution levels varies according to regional economic development levels, with significant locational heterogeneity.

The above analysis is expressed simply in

Table 1. Hypothetical table.

The Literature	Hypothesis	Number
Digital resources and the economy are closely linked Economy interacts with environmental pollution	Favorably reduce the level of environmental pollution	H1
Focusing on the relationship between IS and efficiency	Reduce environmental pollution levels by optimizing IS	H2
Recognizing the potential of digital technologies in energy. Lack of systematic dynamic research	Reduce environmental pollution levels by reducing energy consumption	H3
Low level of quantification of the impact of economic disparities on the development of the digital economy	Impact on environmental pollution varies according to the level of regional economic development	H4

.

3. Research Design

3.1. Modeling

3.1.1. Basic Regression Model

To test Hypothesis 1, the following benchmark regression model is constructed:

$$EP{I}_{it}={\alpha }_0+{\alpha }_{1}ID{E}_{it}+{\alpha }_{2}Contro{l}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

where the dependent variable, EPI (Environmental Pollution Index), represents the level of environmental pollution; IDE denotes the integration of the digital economy and the real economy; Control represents the control variables; i indicates the region, and d denotes time; μ and λ represent the individual and time fixed effects, respectively; and ε is the random disturbance term.

3.1.2. Mediation Effects Model

To test Hypotheses 2 and 3, the mediated effects model [40] was selected for analysis:

$$M_{it}={\beta }_0+{\beta }_{1}ID{E}_{it}+{\beta }_{2}Contro{l}_{it}+{\mu }_{it}+{\eta }_i+{\delta }_t$$

(2)

$$EP{I}_{it}={\gamma }_0+{\gamma }_{1}ID{E}_{it}+{\gamma }_2{M}_{it}+{\gamma }_{3}Contro{l}_{it}+{\epsilon }_{it}+{\mathcal{l}}_i+{\lambda }_t$$

(3)

where M represents the selected mediator variable.

3.2. Selection of Variables

3.2.1. Explanatory Variables

Su et al. (2021) proposed a digital finance index system [42], while Xin et al. (2023) [28] developed an evaluation framework to assess the integration level between the digital economy and industry, employing the entropy-weighted TOPSIS method to measure this integration in China. Meng et al. (2023) [17], based on the input–output approach, derived indicators for the digital economy and real economy within various subsectors of the industrial sector in China and assessed their degree of integration. Chen (2024) analyzed the specific pathways of digital–real integration from three key dimensions—the product level, enterprise level, and industry level—and established a research framework to facilitate the deep integration of the digital–real economy [43]. The indicator systems constructed by these references are shown in

Table 2. System of indicators for the evaluation of the real economy.

Target Level	Standardized Layer	Indicator Layer	Unit (of Measure)	Indicator Properties
Real economy	Scale of development	Gross product	billions	forward
		Gross domestic product (excluding financial real estate)	billions	forward
	Key industry	Financial industry output	billions	forward
		Real estate output	billions	forward
		Total retail sales of consumer goods	billions	forward
	Foreign trade production	Total exports and imports of goods	billions of CNY	forward

and

Table 3. Table of digital economy evaluation indicator system.

Target Level	Standardized Layer	Indicator Layer	Unit (of Measure)	Indicator Properties
Digital economy	Level of digital infrastructure	Internet broadband access port	hundred thousand	forward
		Length of long-distance fiber optic cable lines	kilometer	forward
		Number of domain names	ten thousand	forward
		Number of pages	ten million	forward
		Number of IPv4 addresses	hundred thousand	forward
		Internet penetration	percent	forward
		Cell phone penetration rate	percent	forward
	Level of digital business development	Total postal operations	billions	forward
		Total telecommunication services	billions of dollars	forward
		Express mail volume	million pieces	forward
		Number of patent applications	thousands of items	forward
		R&D expenditures	billions	forward

, with the degree of coupling and coordination between the digital economy and the real economy calculated using the entropy weight method [44].

The entropy weighting method is a widely used objective weighting technique. To understand an unknown entity, the amount of information required for analysis is measured by the quantity of information, and information entropy represents the expected value of this information. The steps and data for the calculation of the entropy weight method can be found in the Supplementary Material.

The coupled coordination model is employed to analyze the level of coordinated development between systems. The degree of coupling refers to the interaction between two or more systems, reflecting the dynamic relationship of coordinated development. The degree of coordination, on the other hand, refers to the extent of beneficial coupling within this interaction, indicating the quality of coordination [45]. An extension of the multi-system coupling model leads to the establishment of a coupled-ness model:

$$C={\left[{\displaystyle \frac{{\displaystyle \prod _i{U}_i}}{{\left(\frac{1}{n}{\displaystyle \sum _i{U}_i}\right)}^n}}\right]}^{{\displaystyle \frac{1}{n}}}$$

(4)

where C represents the coupling degree, ranging from 0 to 1. A value of C = 1 indicates that the system is in a highly coupled state, while C = 0 indicates that the system is in an uncoupled state. Ui denotes the development level of the i-th system.

The coupling degree model is extended to a coupling coordination degree model:

$$T={\displaystyle \sum _i{\alpha }_i\times U_i}(\mathrm{math}.) \mathrm{genus}{\displaystyle \sum _i{\alpha }_{\iota }=1}$$

(5)

$$D=\sqrt{C\times T}$$

(6)

where C represents the coupling degree; D denotes the degree of coupling coordination, which depends on the level of coordinated development between the systems and measures the degree of interdependence and the interconnection of information or parameters within the system; T is the comprehensive reconciliation index between the systems, reflecting the overall synergistic effect; and ɑ_i represents the coefficient to be determined. In this study, ɑ₁ = ɑ₂ = 0.5 is used to emphasize the equal importance of the development of the digital economy and the real economy. The entropy weight method and the coupled coordination degree model are used to derive the IDE score for China.

3.2.2. Explained Variables

The level of environmental pollution refers to the concentration and distribution of harmful substances in the natural environment, such as air, water, and soil, within a specific area. It is typically measured using concentration indicators for various pollutants [46], including harmful gases in the air (e.g., PM2.5, sulfur dioxide, and nitrogen oxides) [47], chemical pollutants in water bodies (e.g., heavy metals, pesticide residues, and organic substances in wastewater), and toxic substances in soil. A single indicator cannot comprehensively and scientifically reflect the overall level of environmental pollution [48]. This study, building on Liu et al. (2019) [49], uses the Environmental Pollution Index (EPI), an abstract, generalized value that comprehensively expresses the level of environmental pollution or environmental quality grade. The EPI is based on total wastewater discharge, sulfur dioxide emissions from exhaust gases, and the quantity of solid, liquid, and gaseous industrial waste produced (

Table 4. System of environmental pollution indicators.

Target Level	Indicator Layer	Unit (of Measure)	Indicator Properties
Environmental pollution emissions	Total wastewater discharge	tons	negative direction
	Sulfur dioxide emissions from exhaust gases	tons	negative direction
	General industrial solid waste generation	tons	negative direction

). The composite score is calculated using the entropy weighting method.

3.2.3. Mediating Variables

The industrial structure index is calculated as the ratio of the output of the tertiary sector to GDP. The tertiary sector, particularly industries such as high-tech, finance, information, education, and other services, typically represents high-quality economic development. As these sectors develop, the overall environmental emission pressure decreases, leading to improved production efficiency and reduced resource waste.

The Energy Consumption Index is based on regional per capita energy consumption, which characterizes the intensity of energy consumption. The integration of the digital economy with the real economy facilitates the development and utilization of new renewable energy sources, diversifies and enriches the energy structure, and reduces reliance on other fossil energy sources.

3.2.4. Control Variables

Based on a comprehensive review of the existing literature [50], and consideration of the actual context, six factors that may influence environmental pollution were selected as control variables [51]: labor productivity, human resources, government intervention, financial development, urbanization rate, and foreign investment [52]. (1) Labor productivity: measured as the ratio of industrial value added to the average number of employees. (2) Human resources: measured by the number of years of education per capita in the region. (3) Government intervention: measured by the ratio of general budget expenditure to GDP. (4) Financial development: measured by the value added of the financial sector relative to GDP. (5) Urbanization level: measured by the ratio of the urban population to the total regional population. (6) Foreign investment: measured by the ratio of total regional foreign investment to GDP [53].

3.3. Data Sources and Descriptive Statistics

Due to limitations in the completeness and availability of raw data, this study utilizes panel data from 30 provincial-level administrative regions in China spanning from 2008 to 2022. The primary data sources include the China Statistical Yearbook, annual statistical yearbooks of various regions, the China Energy Statistical Yearbook, the China Urban Statistical Yearbook, the China Urban Construction Statistical Yearbook, the China Information Yearbook, the China Science and Technology Statistical Yearbook, and the Wind database. Missing data for some provinces and years were imputed using linear interpolation. Descriptive statistics for the variables are presented in

Table 5. Descriptive statistics of variables.

Variable Type	Variant	Description of Indicators	Sample Size	Average Value	(Statistics) Standard Deviation	Minimum Value	Maximum Value
explanatory variable	EPI	Aggregate score	450	0.233	0.142	0.0155	0.672
explanatory variable	IDE	Aggregate score	450	0.304	0.150	0.0617	0.900
intermediary variable	IS	Percentage of third sector	450	0.192	0.141	0.00748	0.762
	EC	Energy consumption per capita	450	4.149	2.413	1.223	11.565
control variable	LP	Value added of industry/average number of persons employed	450	10.12	6.128	1.523	36.76
	GI	General budget expenditure/GDP	450	9.128	0.982	6.764	12.78
	HR	Years of schooling per capita	450	0.110	0.0316	0.0569	0.245
	FD	Financial sector value added/GDP	450	0.0678	0.0318	0.0188	0.197
	UL	Urban population/total population	450	0.582	0.130	0.291	0.896
	FI	Total foreign investment/GDP	450	0.0785	0.0901	0.00141	0.858

.

3.4. Correlation Analysis

A correlation analysis examines the relationships among the elements of a variable. The correlation coefficients were calculated using Stata14, with values ranging from −1 to 1 indicating a significant and strong correlation. Subsequently, the VIF test and Hausman test were conducted, with the average VIF value in the results being less than 10, indicating no issue of multicollinearity in the analysis. The results of the calculations are shown in

Table 6. Correlation analysis results.

	Id	Year	EPI	IDE	IS	EC	LP	HR	GI	FD	UL	FI
id	1
year	0	1
EPI	−0.214 ***	−0.295 ***	1
IDE	−0.302 ***	0.457 ***	0.189 ***	1
IS	0.542 ***	−0.348 ***	0.089 *	−0.599 ***	1
EC	−0.324 ***	0.262 ***	−0.360 ***	0.337 ***	−0.273 ***	1
LP	−0.260 ***	0.599 ***	−0.348 ***	0.550 ***	−0.471 ***	0.614 ***	1
HR	−0.571 ***	0.435 ***	−0.260 ***	0.488 ***	−0.601 ***	0.717 ***	0.704 ***	1
GI	−0.289 ***	−0.0220	−0.252 ***	−0.0130	−0.191 ***	0.356 ***	0.342 ***	0.528 ***	1
FD	−0.250 ***	0.392 ***	−0.500 ***	0.362 ***	−0.425 ***	0.747 ***	0.770 ***	0.748 ***	0.642 ***	1
Ul	−0.545 ***	0.417 ***	−0.301 ***	0.562 ***	−0.689 ***	0.591 ***	0.793 ***	0.877 ***	0.540 ***	0.781 ***	1
FI	−0.308 ***	0.149 ***	−0.280 ***	0.331 ***	−0.508 ***	0.355 ***	0.490 ***	0.540 ***	0.443 ***	0.519 ***	0.629 ***	1

Note: t-values in parentheses, *, and *** indicate significance at 10% and 1% levels.

.

4. Results and Discussion

4.1. Baseline Regression Analysis

Table 7. Benchmark regression results.

	(1)	(2)
VARIABLES	EPI	EPI
IDE	−0.410 ***	−0.500 ***
	(−7.365)	(−8.730)
LP		0.005 ***
		(4.733)
HR		0.025 **
		(2.103)
GI		−0.037
		(−0.222)
FD		−0.303
		(−1.174)
UL		−0.315 ***
		(−3.004)
FI		−0.027
		(−0.692)
Constant	0.342 ***	0.297 ***
	(26.752)	(2.619)
Observations	450	450
R2	0.660	0.699
year FE	YES	YES
id FE	YES	YES

Note: t-values in parentheses, **, and *** indicate significance at 5%,and 1% levels.

presents the results of the baseline regression analysis on the impact of the digital economy on the high-quality development of cities. Two regression models were constructed, one without control variables and the other with all control variables, to examine whether there are differences in the impact of digital–real integration on the Environmental Pollution Index (EPI) and to assess the robustness of the results under different variable specifications.

Column (1) shows the regression results with only the core explanatory variable, the digital–real integration level, included. The regression coefficient for the core explanatory variable, IDE, and the EPI variable is −0.410, which is significant at the 1% level, indicating that a 1% increase in the level of digital–real integration leads to a 0.41% decrease in the EPI. Column (2) presents the regression results after introducing both the core explanatory and control variables. The regression coefficient for IDE and EPI is −0.5, which is significant at the 1% level, indicating that a 1% increase in the level of digital–real integration results in a 0.5% decrease in the environmental pollution composite index.

These results support the argument that digital–real integration plays a key role in optimizing resource allocation and reducing emissions through automation and green innovation. The findings directly address the first research question, which is whether digital–real integration reduces environmental pollution, and confirm its potential as a tool for achieving global sustainable development strategies. Thus, Hypothesis 1 is validated, demonstrating that digital–real integration can reduce the Environmental Pollution Index.

4.2. Robustness Tests

To ensure the reliability of the empirical results, this study refers to the econometric analysis of panel data outlined by Rajarshi Majumder et al. (2020) [54]. The following methods are employed to conduct robustness and endogeneity tests.

(1) Change in sample period

In economics and social science research, the choice of sample can significantly influence the results. This is particularly true for time series data, where a specific time period may possess unique characteristics (e.g., economic crises and policy changes) that could affect the analysis. Therefore, altering the sample period allows for testing the model’s performance across different time frames. Segmenting the analysis provides an opportunity to assess the model’s generalizability and applicability. This robustness test aims to determine whether the model’s conclusions remain consistent under various conditions. If the model yields consistent results across different sample periods, it suggests robustness; however, if the findings differ substantially, it may indicate that the model is sensitive to a particular time period and its conclusions may not be sufficiently reliable. For this test, data from the last ten years (2013–2022) were selected, and the regression results are presented in column (1) of Table 7. It is observed that the regression coefficients for the core explanatory variables, measuring the level of real convergence, pass the significance test at the 1% level, with results largely consistent with those of the benchmark regression.

(2) Replacement of core explanatory variables

To assess the robustness of the benchmark regression results, the sensitivity of the findings to different variable selections or model specifications is tested by replacing the core explanatory variables. If the regression coefficients and significance levels of the modified model align with those of the original model, it suggests that the conclusions are robust. However, substantial changes may indicate that the model is more sensitive to specific explanatory variables, thus compromising the reliability of the results. Based on studies by Gao (2023), Wang (2024), and others on the evaluation of the synergistic effects of pollution and carbon emissions [55], this paper substitutes the carbon emission index as the core explanatory variable. The regression results for environmental pollution levels, with this substitution, are presented in column 2 of

Table 8. Robustness test results.

	(1)	(2)	(3)	(4)
VARIABLES	EPI	EPI	EPI	EPI
IDE	−0.339 ***
	(−3.62)
CE		−0.000 ***
		(−3.42)
IDE			−0.5206 ***
			(0.0000)
L. IDE				−0.466 ***
				(−7.55)
LP	0.003	0.002	0.0062 ***	0.005 ***
	(1.22)	(1.37)	(0.0000)	(4.31)
HR	0.032 **	0.024 **	0.0284 *	0.025 **
	(2.25)	(2.08)	(0.0563)	(1.98)
GI	−0.381	−0.372 **	−0.0146	−0.105
	(−1.48)	(−1.98)	(0.9458)	(−0.61)
FD	−0.437	0.264	−0.1750	−0.174
	(−1.14)	(1.20)	(0.5615)	(−0.63)
UL	−0.645 ***	−0.369 ***	−0.4405 ***	−0.355 ***
	(−2.78)	(−3.47)	(0.0050)	(−3.00)
FI	−0.098	−0.071 *	−0.0099	−0.038
	(−0.84)	(−1.92)	(0.8187)	(−0.72)
Constant	0.478 **	0.324 **	0.3298 **	0.308 **
	(2.30)	(2.53)	(0.0117)	(2.45)
Observations	300	450	390	420
R-squared	0.946	0.933	0.6647	0.705
Year FE	Yes	Yes	Yes	Yes
id FE	Yes	Yes	Yes	Yes

Note: t-values in parentheses, *, **, and *** indicate significance at 10%, 5%, and 1% levels.

. Although replacing the core explanatory variable leads to significant changes in the results, with values approaching zero, the significance levels remain, confirming that the core explanatory variable continues to be statistically significant [56].

(3) Culling samples

Considering that certain cities may exhibit distinct advantages or disadvantages due to factors such as policy favoritism, it is important to acknowledge that municipalities, as cities with significant location, economic, and political advantages within China’s development plan, tend to have greater development opportunities. Consequently, these cities may require the consideration of more factors in their urban development, which could differ from the development trends observed in regular provincial units. Therefore, to test the robustness of the previous regression results, this study performs the regression again after excluding the four municipalities from the thirty provincial administrative units. The corresponding regression results are shown in column 3 of Table 7. It is evident that the regression coefficients for the core explanatory variables related to real convergence pass the 1% significance test, and these results are largely consistent with those of the benchmark regression.

(4) Endogeneity test: lagged one period of the independent variable

In this test, the current value of the independent variable is replaced by its value from the previous period (or earlier), a method commonly used in dynamic effects analysis. In certain contexts, phenomena may exhibit delayed effects, meaning the current dependent variable may be influenced not only by the current independent variable but also by its past values. This can create endogeneity issues between the independent and dependent variables, such as bidirectional causality and omitted variable bias. On one hand, regions with initially high economic levels may be more inclined to develop a green economy; on the other hand, the historical level of environmental pollution in each region often serves as a reference point for government bodies when formulating development strategies. These endogenous issues may result in biased and unreliable estimates. By introducing lagged values of the independent variable, the impact of endogeneity problems, such as two-way causality, can be mitigated. This approach enhances the robustness of the model and helps reveal the delayed effects. The results, presented in column 4 of Table 8, indicate that the core explanatory variables remain significant at the 1% level, consistent with the benchmark regression results.

(5) Endogeneity tests: instrumental variables

To address potential endogeneity issues in the econometric model, this study follows the approach of Pang et al. (2025) [57] and uses the historical number of post offices as an instrumental variable. This choice is justified by the following reasons: The development of Internet technology in China began with the widespread adoption of fixed-line telephones. Areas with a higher historical penetration of fixed-line telephones later exhibited higher Internet penetration. Post offices, being the executing agencies for the installation of fixed-line telephones, played a key role in the distribution and development of telephone networks. As a result, the historical distribution of post offices indirectly influenced the spread of the Internet, whereas it does not directly affect the Environmental Pollution Index (EPI). Therefore, the historical number of post offices meets the relevance and exogeneity conditions required for an effective instrumental variable.

Specifically, the number of post offices in each province in 1990 remained constant over time. To make the variable dynamic, the interaction term (IV), derived by multiplying the investment in fixed assets for information transmission, computer services, and the software industry in the previous year, is used as the instrumental variable for the digital economy. After passing tests for weak instrument validity and the sufficiency of instrumental variables, the contribution of the IDE to the improvement of the EPI remains robust. The results of the calculations are shown in

Table 9. Instrumental variable regression.

VARIABLES	EPI
	(1)2SLS	(2)2SLS
IDE		−0.412 ***
		(−6.32)
IV	−0.021 ***
	(−0.13)
Mediating Variables	YES	YES
year FE	YES	YES
id FE	YES	YES
Kleibergen—Paap rk LM		53.24 ***
Kleibergen—Paap rk Wald F		61.45

Note: t-values in parentheses, *** indicate significance at 1% levels.

.

4.3. Analysis of Locational Heterogeneity

China’s vast land area and uncoordinated regional development present significant challenges to overall progress. Differences in geographic location can lead to varying degrees of integration between the real and digital economies, which, in turn, affect environmental pollution levels [58]. Zhang et al. (2023) examined the impact of the digital economy on environmental quality using panel data from China, revealing significant regional heterogeneity. The western region showed the strongest inhibitory effect, followed by the eastern region, while the central region exhibited the weakest impact. This finding suggests that the relationship between regional economic development and environmental pollution is not necessarily linear [59].

To account for regional variations, the data were divided into eastern, central, and western regions based on the National Statistical Office’s criteria.

Table 10. Robustness test results.

	(1)	(2)	(3)
VARIABLES	EPI	EPI	EPI
IDE	−0.464 ***	−0.390 **	−0.758 ***
	(−4.98)	(−2.18)	(−5.64)
LP	0.006 **	0.002	0.003
	(2.43)	(0.31)	(1.29)
HR	0.035 **	0.022	0.002
	(2.00)	(0.88)	(0.11)
GI	0.754 **	−0.574 **	−0.535 **
	(1.98)	(−2.16)	(−2.04)
FD	−0.265	−0.060	−0.206
	(−0.80)	(−0.10)	(−0.72)
UL	−0.089	−0.751	−1.804 ***
	(−0.64)	(−1.24)	(−6.49)
FI	−0.030	0.079	0.211 **
	(−0.68)	(0.28)	(2.35)
Constant	0.018	0.695	1.274 ***
	(0.10)	(1.30)	(6.07)
Observations	195	120	135
R-squared	0.944	0.963	0.932
year FE	Yes	Yes	Yes
id FE	Yes	Yes	Yes

Note: t-values in parentheses, **, and *** indicate significance at 5% and 1% levels.

presents the results of the regional heterogeneity regression. It shows that the regression coefficients for the IDE in the eastern provinces passed the significance test at the 1% level, those for the central provinces passed at the 5% level, and those for the western provinces passed at the 1% level. The results confirm that the development of IDE can effectively reduce environmental pollution, with the impact being more pronounced in the western regions, followed by the eastern and central regions. These findings underscore the importance of region-specific policies, a gap that has been overlooked in previous studies.

As the forefront of China’s economic development, the eastern region enjoys significant advantages in industrial innovation, foreign trade, and economic cooperation. The digital economy here started earlier, with the deeper penetration and broader application of digital technologies, allowing it to fully leverage digital economy dividends. However, the region’s high pollution emissions base limits the overall reduction effect.

In contrast, central China has served as a key area for the transfer of eastern industrial development. Although there is a development gap compared to the eastern region, it has effectively addressed shortfalls in digital technologies such as 5G and the industrial internet through industrial transfers. Despite these advancements, traditional industries still dominate in this region, and the emission reduction effect from IDE remains relatively moderate.

The western region, with its lower level of economic development and less-developed industrial structure, has fewer traditional high-pollution, high energy-consuming industries. As digital industries develop, the region has the potential to achieve a green, low-carbon transformation in a relatively short period. The emission reduction effect of the IDE is more pronounced in this region, particularly given the recent emphasis on ecological protection and improved policy support in the western region.

4.4. Analysis of Intermediary Mechanisms

4.4.1. Industrial Structure

The regression results mediated by the industrial structure indicate that the regression coefficient for the level of digital–real integration passes the significance test at the 1% level. This suggests that digital–real integration can significantly enhance the rationalization of the industrial structure in various regions. The regression coefficient for the industrial structure also passes the significance test at the 1% level, providing preliminary support for Hypothesis 2 of this study.

In the process of IDE promoting the reduction in environmental pollution, a rational and advanced industrial structure plays a crucial intermediary role. It facilitates the transformation of traditional high-pollution industries into low-carbon green industries. This transformation not only improves the efficiency of the production process and promotes green technological innovation but also strengthens synergies. Additionally, the support of green finance and the shifts in consumer demand further accelerate industrial transformation, leading to comprehensive pollution reduction. The results are shown in

Table 11. Table of intermediary mechanisms of industrial structure.

	(1)	(2)	(3)
	EPI	IS	EPI
IDE	−0.560 ***	26.859 ***	−0.457 ***
	(−9.491)	(7.802)	(−7.405)
LP	0.004 ***	−0.321 ***	0.003 **
	(3.642)	(−4.766)	(2.566)
HR	0.037 ***	−0.141	0.037 ***
	(4.039)	(−0.260)	(4.080)
GI	0.923 ***	−62.194 ***	0.683 ***
	(6.485)	(−7.495)	(4.618)
FD	−0.524 **	192.062 ***	0.216
	(−2.450)	(15.402)	(0.825)
UL	−0.478 ***	28.927 ***	−0.367 ***
	(−6.448)	(6.691)	(−4.814)
FI	−0.091 **	0.306	−0.089 **
	(−2.090)	(0.121)	(−2.115)
IS			−0.004 ***
			(−4.678)
_cons	0.238 ***	21.540 ***	0.321 ***
	(3.729)	(5.799)	(4.959)
N	450	450	450
R2	0.605	0.823	0.625
F	90.401	273.753	85.837
year FE	YES	YES	YES
id FE	YES	YES	YES

Note: t-values in parentheses, **, and *** indicate significance at 5% and 1% levels.

4.4.2. Energy Consumption

The regression results with EC as the mediator indicate that the regression coefficient for IDE passes the significance test at the 1% level, demonstrating that IDE significantly enhances energy consumption efficiency across regions. Additionally, the regression coefficient for EC also passes the significance test at the 1% level, supporting Hypothesis 3 of this study.

Energy consumption plays a key intermediary role in the reduction in environmental pollution through IDE. By applying digital technologies, efficiency can be improved, reducing ineffective energy consumption. For example, intelligent and precise production scheduling enables better energy control, preventing excessive consumption. Simultaneously, digitalization fosters the research, development, and application of green energy, facilitating the substitution of clean energy and optimizing the overall energy structure.

These findings explicitly test the second and third research hypotheses, shedding light on how (rather than just whether) the integration of the digital and real economies contributes to pollution reduction. The results are shown in

Table 12. Table of intermediary mechanisms for energy consumption.

	(1)	(2)	(3)
	EPI	EC	EPI
IDE	−0.560 ***	70,201.604 ***	−0.325 ***
	(−9.491)	(20.825)	(−3.909)
LP	0.004 ***	176.131 ***	0.005 ***
	(3.642)	(2.669)	(4.190)
HR	0.037 ***	0.808	0.037 ***
	(4.039)	(0.002)	(4.111)
GI	0.923 ***	−1.66 × 104 **	0.867 ***
	(6.485)	(−2.041)	(6.171)
FD	−0.524 **	−4.01 × 104 ***	−0.658 ***
	(−2.450)	(−3.283)	(−3.093)
UL	−0.478 ***	−4.26 × 104 ***	−0.621 ***
	(−6.448)	(−10.068)	(−7.639)
FI	−0.091 **	3390.610	−0.079 *
	(−2.090)	(1.370)	(−1.856)
EC			−0.000 ***
			(−3.964)
_cons	0.238 ***	9467.203 ***	0.269 ***
	(3.729)	(2.603)	(4.268)
N	450	450	450
R2	0.605	0.741	0.620
F	90.401	168.857	83.882

Note: t-values in parentheses, *, **, and *** indicate significance at 10%, 5%, and 1% levels.

.

4.5. Nonlinearity Test

The threshold effect denotes the phenomenon whereby changes in a given variable or condition produce significant effects only after a critical threshold is reached. Thus, the relationship between variables is not linear or continuous but exhibits a specific threshold beyond which the effect becomes pronounced [60].

The threshold effect demonstrates that the influence of certain variables is gradual, with no significant changes occurring until a “critical value” is surpassed. This phenomenon is common across various fields, including economics, environmental studies, and sociology. Understanding the threshold effect can help reveal the nonlinear relationships between changes in variables and outcomes in complex systems, and it has been extensively studied in both mathematical and practical contexts [61].

In this study, industrial structure and energy consumption are used as threshold variables, with 300 iterations of sampling performed. The results of the threshold effect test are shown in

Table 13. Threshold effects.

Threshold Variable	Threshold Number	Threshold Value	p-Value	95% Confidence Interval		Number of BS
				Lower Limit	Limit
Industrial Structure	single threshold	46.8000	0.023	46.3500	47.2000	300

, indicating the presence of a single-threshold effect in the industrial structure.

Referring to the descriptive analysis of Wang et al. (2025) [62] regarding the threshold effect, the results indicate that the regression coefficient of the EPI shifts from −0.271288 to −0.4094245 once the industrial structure (IS) surpasses the threshold. In the context of the gradual development of the IDE level, IS demonstrates a significant threshold effect on EPI. The optimization of the industrial structure has a notably stronger impact on pollution reduction after crossing a specific threshold of 46.8, with the effect significantly enhanced.

This may be attributed to the fact that as IDE rises, digital technologies and intelligent methods progressively infiltrate various industrial sectors. When the shift toward a more efficient and environmentally friendly direction begins, its effect on environmental pollution does not immediately manifest as a substantial change. Instead, it gradually accumulates until a critical point is reached, at which point the reduction in pollution becomes more pronounced.

This threshold effect indicates that the process of industrial structure optimization requires specific foundational conditions and time accumulation. Moreover, the application of intelligent resource allocation and green technology can significantly reduce the environmental impact.

5. Conclusions and Implications

5.1. Conclusions

This study investigates the effect of the integration level of digital and real economies on environmental pollution in China and explores the underlying mechanisms. Using the two-way fixed-effects model, the panel mediation model, and the threshold model, we conduct empirical analyses on data from 30 provincial-level administrative regions in China (excluding Hong Kong, Macao, Taiwan, and the Tibet Autonomous Region) for the period 2008–2022. The following conclusions are drawn:

1. The integrated development of digital and real economies reduces environmental pollution. This result is consistent with various robustness checks.

2. Heterogeneity analysis reveals regional variation in this effect. It is stronger in western regions than in eastern and central regions, reflecting differences in local economic and ecological foundations.

3. Mechanism analysis shows that industrial structure optimization and energy consumption reduction are the key channels through which digital–real integration improves environmental quality. Moreover, the industrial structure variable exhibits a threshold at 46.8. Above this value, the influence of digital–real integration on environmental pollution is nonlinear, with an increasing marginal effect.

These findings align with global evidence. For example, the EU practice of reducing carbon emissions through digital technologies suggests that the environmental benefits of digital–real integration may be universal. Similar regional disparities appear in developing countries, such as the digital divide in India, highlighting the need for region-specific policies. Therefore, our results have cross-regional relevance and may inform the global sustainability transition.

5.2. Theoretical and Practical Implications

Theoretical significance:

1. Expands the research framework on the environmental effects of digital–real integration by verifying the dual mediating roles of the industrial structure and energy consumption, providing a new explanatory mechanism for environmental economics.

2. Reveals regional heterogeneity and nonlinear effects, challenging the traditional linear policy-design logic and emphasizing the need for more studies on policy optimization under threshold conditions.

Practical implications:

Suggestions for China’s policies:

1. Strengthen the application of mediating and threshold effects by focusing on industrial upgrading and digital transformation in the energy sector. Provinces with an industrial structure index below 46.8 (e.g., Henan and Anhui) could focus on the digitization of traditional manufacturing industries (IS = −0.004, Table 11), while regions above this threshold (e.g., Guangdong and Zhejiang) could encourage an AI-driven circular economy in high-tech industries.

2. Implement region-specific governance strategies: Given the stronger inhibitory effect of IDE on pollution in western China (−0.758 inhibition greater than −0.464 in the east, Table 10), the region could prioritize 5G infrastructure and rural e-commerce platforms to extend the environmental benefits of IDE. The eastern region should focus on green technology innovation, reduce energy consumption by expanding the adoption of e-commuting infrastructure, for example, in eastern cities (EC = −0.000, Table 12); and the central region should strengthen cross-regional synergies through data-sharing platforms.

Implications for developing economies:

1. Promote digital integration in phases: countries with weaker digital economies should pilot digital tools in key sectors such as agriculture and energy, and then gradually expand their scope.

2. Balance efficiency and equity: drawing on China’s experience with digital poverty alleviation, governments should narrow the regional digital divide through financial subsidies and technology transfers, avoiding environmental–governance policies that exacerbate social inequality.

Direction of international cooperation:

1. Establish a global database on the environmental impacts of transnational digital convergence, promoting the worldwide adoption of standardized indicators such as the Green Digital Economy Index.

2. Facilitate knowledge sharing among developing countries on digital environmental technologies, for example, blockchain-based carbon tracking, through South–South cooperation mechanisms.

5.3. Research Limitations and Outlook

Although the research findings offer certain theoretical and practical value, this study has the following limitations:

1. Limited data coverage. The analysis relies on provincial panel data and omits firm- or industry-level microdata. Consequently, internal heterogeneity may be overlooked, and measurement errors may arise. Moreover, the use of composite indices (for example, EPI and IDE) may oversimplify complex interactions. Although these indices aid operationalization, they may obscure important regional differences in the pollutant composition or sector-specific adoption rates of digital technologies, which are critical for formulating tailored policies.

2. The insufficient consideration of counterfactuals. This study does not fully account for potential counterfactual scenarios. For example, in provinces where digital technologies are used to optimize fossil fuel extraction or energy-intensive manufacturing (for instance, Inner Mongolia), the environmental benefits of digital–real integration may be offset by increased resource extraction. This issue challenges the generalizability of the proposed mechanism and warrants further investigation in future research.

3. A lack of international comparisons. The conclusions are drawn primarily from the Chinese context. The applicability and relevance of these findings to other countries, particularly developing economies with weaker digital economy foundations, require further validation.

4. The absence of dynamic impacts. This study does not examine the long-term dynamic effects of digital–real integration. Incorporating time-series analyses and panel data spanning multiple economic cycles would strengthen causal inferences and provide a basis for iterative policy development.