Can the digital economy development achieve the effect of pollution reduction? Evidence from Chinese Cities

Abstract

As a new economic form, the digital economy is not only empowering new impetus to economic growth, but also reshaping specific business forms of economical operation. Therefore, we conducted an empirical test to verify the impact and mechanism of pollution reduction in the digital economy, based on the panel data of 280 prefecture-level cities in China from 2011 to 2019. The results show that, first the development of the digital economy indeed has the positive effect of realizing pollution reduction. The results of mediating effect test indicate the influence mechanism mainly rely on promoting the upgrading of industrial structure (structural effect) and upgrading the level of green technology innovation (technical effect). Second, the results of regional heterogeneity analysis show that the emission reduction effect of digital economy development on four pollutants is characterized by weakness in the east and strong in the west in regional distribution. Third, the development of digital economy has a threshold effect on the level of economic development to achieve its pollution reduction effect. Further identification of the threshold effect indicates that the higher the level of economic development, the better in emission reduction effect.

Introduction

Industrialization and urbanization are considered as crucial ways for a country or region to realize modernization and economic growth (Xiao et al. 2022). However, it also has caused severe environmental pollution problems, posing challenges to the sustainable development of the global economy (Xu et al. 2020; Mesagan et al. 2022). Benefiting from the Reform and Opening-up policy, China's economy has experienced rapid development over the past 40 years, creating a rare miracle of economic growth in the world (Sheng et al. 2020). However, the GDP championship of local governments has made environmental protection places a low priority in policy choices for a long time, and the extensive development model has led to the over-consumption of resources and the massive emission of pollutants (Wen et al. 2021; Irfan et al. 2021). Environmental pollution not only brings irreparable ecological trauma to China (Zhang et al. 2020), but also has become a prominent weakness in the process of achieving high-quality development. Although the central government of China has issued a series of environmental protection policies and measures to eliminate the negative impact of environmental pollution, such as collecting environmental taxes, promoting environmental supervision, strengthening environmental regulations, etc., the reality is that the total amount of environmental pollution driven by economic growth is still increasing. In general, the prospect of environmental governance in China is not optimistic (Li et al. 2022a). According to the Global Environmental Performance Index (EPI) Report 2022, jointly released by Yale University and other research institutions, China's environmental performance index only ranks 160th among 180 countries and regions with a score of 28.4 (Lyu et al. 2023), which further betokens that environmental pollution control still is the one of focus for the Chinese government in the future (Zhang et al. 2020).

The digital economy was defined by OECD as a series of economic activities with digital knowledge and information as crucial production factors, modern information networks as an essential carrier, and the effective use of information and communication technologies as a main driving force for efficiency improvement and economic structure optimization (OECD 2014). As a new economic form, the digital economy is penetrating all aspects of social and economic life and reshaping the global socio-economic structure (Zhu et al. 2022). With the continuous expansion of the breadth and depth of digital technologies such as the Internet, cloud computing, the Internet of Things, and blockchain, the digital economy has gradually become one of the most dynamic industries in the global economy (Pan et al. 2022). The White Paper on the Development of the Global Digital Economy (2022) issued by the China Academy of Information and Communication Technology pointed out that, the scale of the added value of the digital economy in 47 countries around the world reached 38.1 trillion US dollars in 2021, with nominal year-on-year growth of 15.6%. Meanwhile, as one of the biggest promoters and beneficiaries of the development of the digital economy, China's digital economy scale reached 7.1 trillion US dollars in 2021, ranking second in the world, increasing by 31.48% over 2020, which showed the vast development potential. Against the backdrop of the exogenous impact of COVID-19 and the sluggish global economic growth, China builds her hopes on the development of the digital economy will serve as an essential driving force for future economic growth to alleviate the current difficulties on economic operation, such as the diminishing marginal returns on factors and the urgent transformation of new and old kinetic energy (Wu et al. 2021). In the Outline of the People's Republic of China 14th Five-Year Plan for National Economic and Social Development and Long-Range Objectives for 2035, it is clearly stated that China will embrace the digital era, unlock the potential of big data, accelerate the development of the digital economy, digital society, and digital government, and transform the pattern of production, lifestyle, and governance models through digital transformation. In this context, whether the digital economy development achieve the effect of pollutants reduction, and then bring about environmental welfare effects for China's growth? If so, what is the internal mechanism that the digital economy achieves pollution reduction? The discussion of the above issues is not only conducive to better promote the development of the digital economy, but also has practical significance for promoting the improvement of the ecological environment and building ecological friendly China.

Literature review

Currently, there are two research branches highly relevant to this article. As for the influencing factors of environmental pollution, the Environmental Kuznets Curve (EKC) proposed by Grossman and Krueger (1995) focused on the relationship between economic growth and environmental protection. It was believed there was an inverse nonlinear relationship in U-shape type between per capita income and environmental pollution level. However, some scholars hold different opinions about EKC, believing that there is still a single linear relationship between economic growth and environmental pollution (Chen et al. 2020; Rao and Yan 2020). In addition, some scholars claimed that there should be other types of nonlinear curve relationships, such as U-shaped (Smulders et al. 2011) and inverted N-shaped curves (Li et al. 2022b). With the acceleration of globalization, the relevant research perspective gradually turned to the environmental pollution effect of foreign direct investment, the theory hypothesis such as pollution refuge (Levinson and Taylor 2008; Khan and Ozturk 2020) and pollution halo (Letchumanan and Kodama 2000) emerged. Adeel-Farooq et al. (2021) examined the impact of FDI on the overall environmental quality of 76 countries, found that FDI from developed improved the environmental performance of low-income and high-income host countries at same time, supporting the pollution halo hypothesis. Tian et al. (2023) based panel data of thirty-seven industrial sectors in China from 2003 to 2015, founds the coordinated development of inward and outward FDI has a positive impact on reducing pollutants emissions. Further research indicates that there is no contradiction between the pollution paradise and the pollution haven hypothesis, and the key to explaining this issue lies in the environmental regulation, technical level and institutional quality of the host country (Huynh and Hoang 2019; Wang and Luo 2020; Yoon and Heshmati 2021). In recent years, it gradually becomes a hot topic about the impact of transportation infrastructure on environmental pollution. For example, Yang et al. (2019) took sulfur dioxide emissions per unit output as the proxy variable of pollution and found that it would reduce local pollution emissions by 7.35% when the high-speed railway put into operation. Based on differences-in-differences model, Gao et al. (2021) found that high-speed train commenced operations reduced the pollution emissions from heavy pollution industries, shed the air pollution in central cities and urban central areas. Besides, partial literatures exploring the impact factors of environmental pollution from the perspectives of urbanization (Rahman and Allam 2021; Xu et al. 2022), environmental regulation (Yang et al. 2018; Lu et al. 2021), fiscal decentralization (Zhao et al. 2022; Guo et al. 2020a), and technological innovation (Wang and Wang 2022).

As a new economic form, a series of changes caused by the rise of the digital economy has attracted extensive attention from scholars. The research concerning the digital economy mainly focuses on the measurement and evolution characteristics of digital economy index (Ojanperä et al. 2019), the impact of digital economy on green total factor productivity (Liu et al. 2022a; Lyu et al. 2023), high-quality economic development (Lu and Zhu 2022; Li et al. 2022c), technological innovation (Wang et al. 2022a) and industrial structure (Su et al. 2021). As an important branch, scholars inquired into the relationship between the digital economy and environmental pollution from different perspectives. The positive effect of the digital economy in reducing emissions of various pollutants has been widely confirmed in relevant research. Such as Che and Wang (2022) found there is a significant negative correlation between the development of the digital economy and haze pollution. Li et al. (2022a) found development of the digital economy has dramatically reduced PM2.5 emission level in China. Furthermore, the direct positive impact of the digital economy on other single pollutants has also been confirmed. Research by Wang and Chen (2022) indicates that the development of the digital economy has promoted the industrial upgrading process of resource-based cities in China, thereby reducing local air pollution. Wan and Shi (2022) validated that imposing reasonable environmental regulations can more effectively play the role of the digital economy in reducing environmental pollution. On the other hand, some studies have noted the spillover effects of the development of the digital economy. Such as Zha et al. (2022) found that the development of digital economy can not only abate the carbon emission intensity of the specific region, but also effectively improve the environmental performance of surrounding cities. Shen et al. (2022) found that the development of local digital economy is highly correlated with economic development level, and the development of digital economy has a positive spillover effect on environmental performance in surrounding areas. In addition, the mechanism by which the digital economy can promote pollutant reduction also is hot issue in this research field. The existing research mainly focuses on the effects of technological progress and industrial structure (Hu and Guo 2022; Yang et al 2023; Zhao et al. 2023). Such as Zhou et al. (2021) believed that the digital economy can reduce haze pollution via the path mechanism of enhancing the proportion of advanced industrial structure in GDP. Sun et al. (2022) confirmed that the digital economy promoting innovation is an important mediation mechanism that affects PM2.5 pollution through the mediation effect model.

Overall, although many studies have explored the relationship between the development of the digital economy and environmental pollution, the existing literature focuses on the perspective of specific single pollutant. In view of this, based on the panel data of 280 cities at prefecture-level in China from 2011 to 2019, this study constructed multiple econometric models such as fixed effect, mediating effect and threshold effect to explore whether the development of the digital economy can achieve the reduction effect of variety of pollutants, and further explore its influence mechanism, heterogeneity and threshold effect of economic development. The possible marginal contributions of this paper may include the following aspects: Firstly, combined the theory and empirical research, this paper discusses the emission reduction effect of various pollutants in the digital economy at the same time, which enriches stylized facts and research about the impact of the digital economy to environmental pollution. Secondly, to better solve the endogenous problem, we take the Broadband China strategy as an important quasi natural experiment in the development of China's digital economy, and estimate the impact on various environmental pollutants using the differences in differences (DID) model, which provides a more reliable empirical reference for enabling environmental pollution control of the digital economy. Thirdly, by introducing the economic development level into the analysis framework, we used the threshold model to explore the coordination and adaptation of the economic development level with the digital economy in the process of realizing the emission reduction effect, thereby providing a new perspective for better releasing the pollutant emission reduction potential of the digital economy.

Research hypothesis

Digital economy and environmental pollution

The existing research generally believes that the essence of sustainable development is to achieve the coordinated development of economy, society, resources and environment, and digital economy based on modern information and communication technology provides a new opportunity to realize pollutant emission reduction and sustainable development (Zhang et al. 2022a; Han and Liu 2022). Specifically, the impact of digital economy development on environmental pollution is mainly reflected in four aspects: Firstly, the penetration of the digital economy within enterprises contributes to the construction of enterprise green production mode, achieving the source control of pollutants. With the development of the digital economy, enterprises as the main body of pollution prevention can introduce cloud computing, virtual reality and other digital technologies to simulate the production process, effectively integrate all kinds of resources in the production process, alleviate the fragmentation and asymmetry of information, improve energy utilization efficiency, reduce unnecessary energy consumption and pollutant emissions (Zhan et al. 2018). Through the integration of the digital technology and own production mode, enterprises have also achieved optimization in their previous production processes, promoting their development towards a more environmentally friendly direction. On the other hand, the development of the digital economy has also spawned many emerging industries. Compared to traditional industries, the digital economy itself has the charactered by lower pollutant emissions and better economic driving effect. Second, with the continuous development of economic scale and the increasing complexity of ecological and environmental problems, the traditional ecological supervision model is gradually inapplicable to China. The development of the digital economy coincides with the demand that the government optimizes the environmental supervision model and provides technical support and assistance. In fact, China has adopted a large number of new digital technology in the field of environmental protection to improve its own environmental monitoring, such as building a comprehensive range of environmental monitoring points to collect primary data, and combining big data, cloud computing, and remote sensing technology to dynamically monitor air quality, river water quality, pollution emissions, environmental carrying capacity, etc. (Hampton et al. 2013; Shin and Choi 2015). The above initiatives by the Chinese government dramatically upgrade the early warning and perception of pollution sources, enhance the accuracy and timeliness of environmental supervision, and ameliorate the government environmental supervision level further. Ultimately, it forces high polluting enterprises to continuously reduce their pollutant emissions in order to keep up with increasingly strict environmental regulations in China. Third, in the context of the development of the digital economy, digital media has built a new bridge for information exchange and sharing between the government and society. The development of the Internet has made environmental protection concepts and information spread faster and more widely. More people get environmental information through digital media, which objectively enhance public environmental concern and perceptions (Schmidthuber et al. 2019). And the new media platforms such as Weibo, WeChat, and TikTok provide convenient environmental supervision channels for the public, which is conducive to the public reporting corporate pollution behavior to the government and improving the efficiency of social environmental protection supervision (Yang et al. 2020). Fourth, in terms of residents' lives, the digital economy has promoted the prosperity of home office, telemedicine and online education. the new life and working methods will directly reduce energy consumption and automobile exhaust emissions caused by commuting and idle resources. Therefore, the first hypothesis is proposed:

• H1: The development of the digital economy is conducive to reducing the emission of various environmental pollutants.

Influence mechanism

As stated in the literature review, the emission reduction effect of digital economy on pollutants can be realized mainly through two paths: structural effect and technical effect. Among them, the structural effect mainly is shown as the upgrading of industrial structure. The technological effect can be expressed as the green technical innovation (Grossman and Krueger 1995; Jalil and Feridun 2011).

Specifically, on the one hand, the wide application of digital technology will promote the development of digital industry such as communication services, information technology, software, and the Internet. The emerging industries tend to use pollution-free and clean production factors for production, and the environmental pollution caused by these emerging digital industries is far lower than traditional industries while driving economic development. The digital economy will also spawn new business models (such as online shopping), which can drive the formation of new industries. Thus, the rise of emerging industries has drastically changed the industrial structure, further increasing the proportion of the tertiary industry in GDP. On the other hand, the digital economy also can break the boundaries between sectors, reshape the ecological rules of traditional industries, promote the combination of new production factors and new infrastructure. The development of the digital economy makes high pollution and high energy consumption industries eliminate gradually, and strategic emerging industries and modern service industries accelerate their development. As a result, the pollutant emissions in economic development have decreased during the alternation of new and old industries (Liu et al. 2022b). At the same time, the development of the digital economy has also triggered transformation in traditional manufacturing production. Data elements included in the digital economy can accelerate the deep integration of the digital economy and traditional sectors, improve operation efficiency, and promote the transformation and upgrading of conventional industrial structures through industrial linkages, technology diffusion and other effects (Lyu et al. 2023). Faced with fierce market competition, traditional manufacturing enterprises that actively introduce the digital economy are able to cultivate new competitive advantages, adjusts the production mode, and introduces more advanced and environmentally friendly production technology and pollutant end treatment equipment, so as to greatly reduce the pollutant output and emissions in the production process.

The development of the digital economy not only promotes the change of economic structure, but also accelerates the penetration of technological innovation into industries. For instance, the application of big data, cloud computing and other digital technologies is applicating in searching, integrating, analyzing, and making decisions on green product information and green consumption preference information. The enterprise can judge the potential, direction, and path of green technological innovation through the integration of consumer data, to promote manufacturers' innovation from experience driven to data-driven, thus reducing innovation costs (Johnson et al. 2017). The digital economy also will strengthen the diffusion effect of digital technology, accelerate the green technology spillover to other industry sectors, and promote the digitalization and green transformation of traditional enterprises. The technological innovation brought by digital economy is more inclined to green, energy saving and emission reduction, which is conducive to promoting green technology innovation and reducing pollution emissions. In response to the tremendous technological changes brought about by the development of the digital economy, the enterprises will force the R&D and application of clean technology, promote the formation of raw material procurement, green product production and transformation, intelligent logistics warehousing and sales circulation based on digital technology (Ning et al. 2022). In the process, enterprises also can effectively promote the intellectual and flexible transformation of enterprises, gradually change the original energy consumption mode in business, reduce redundancy and intermediate consumption in the production process, and promote enterprise pollution reduction. Therefore, this study proposes the following assumptions:

• H2: The digital economy can reduce environmental pollution emissions by promoting the upgrading of industrial structure (structural effect) and upgrading the level of green technology innovation (technical effect).

Model design and variable description

Model Setting

To test the effect the digital economy on pollutant reduction, the following benchmark model is built:

$$ln{Y}_{\mathit{it}}={\alpha }_0+{\alpha }_{1}ln{\mathit{de}}_{\mathit{it}}+\alpha X_{\mathit{it}}{+\epsilon }_{\mathit{it}}$$ 1

where, subscripts $i$ and $t$ represent city and year respectively.$Y_{\mathit{it}}$ is the environmental pollution discharge amount of $i$ city in year $t$, including PM2.5 ($pm2.5$), industrial wastewater discharge ($\mathit{water}$), industrial sulfur dioxide emission ($so2$), industrial soot and dust emission ($\mathit{dus}$ t). The variable $\mathit{de}$ indicates the development level of the digital economy of city $i$ in year $t$. ${\alpha }_1$ is estimated coefficient of core explanatory variable, which describes the impact of the digital economy on environmental pollutants. $X_{\mathit{it}}$ represents the set of control variables, including city size ($\mathit{ps}$), R&D investment ($\mathit{tc}$), urbanization level ($\mathit{ur}$), government intervention ($\mathit{gov}$), foreign direct investment ($\mathit{fdi}$) and transportation infrastructure level ($\mathit{inf}$). ${\mathrm{\epsilon }}_{\mathit{it}}$ is a random error term.

To further explore the influence mechanism of the digital economy to achieve pollutant reduction, concerning Baron and Kenny (1986)'s step-by-step regression method of mediating effect, the following mediating model is built based on formula (1):

$$ln{\mathit{Med}}_{\mathit{it}}={\beta }_0+{\beta }_{1}lnd{e}_{\mathit{it}}+\beta X_{\mathit{it}}{+\epsilon }_{\mathit{it}}$$ 2

$$ln{Y}_{\mathit{it}}={\lambda }_0+{\lambda }_{1}ln{\mathit{de}}_{\mathit{it}}+{\lambda }_{2}ln{\mathit{Med}}_{\mathit{it}}+\lambda X_{\mathit{it}}{+\epsilon }_{\mathit{it}}$$ 3

where, ${\mathit{Med}}_{\mathit{it}}$ expresses the mechanism variable, that is, industrial structure upgrading (is) and green technology innovation (gp) respectively. The meaning of other variables is same as Eq. (1). Equation (1)—Eq. (3) constitutes the classical mediating effect model. According to Baron and Kenny (1986)'s the inspection idea, if only coefficient ${\alpha }_1$ passes the significance test in Eq. (1), the next step of the mechanism test analysis can be carried out. The second step of the mediating effect test need to verify the impact of the core explanatory variable on the mechanism variable (i.e., Eq. 2). When estimated coefficient ${\beta }_1$ is positive and significant in Eq. (2), it shows that the digital economy and mediating variables changes in the same direction, thereby establishing a significant statistical correlation between mediating variables and core explanatory variable. The final step is to test the coefficient of λ₂ in Eq. (3). When λ₂ also passed the significance test in Eq. (3), mediating effect takes hold, and β₁*λ₂ indicates the size of the mediating effect.

Main variables

1. Explained variable. At present, there is still no consensus on the unified approach of measure environmental pollution. Some scholars integrate multiple pollutants into a single indicator by mathematical method (Liao 2018; Bai et al. 2022), while others take single pollutants emission as a proxy variable of environmental pollution (Wu et al. 2023; Amin et al. 2022; Qi et al. 2022). Although the comprehensive index of environmental pollutants gives due considerations of the diversity of pollutants, the weighted method used during synthesizing a single index makes it more reflected in the relative pollutant emission level of contaminants rather than absolute emissions. Therefore, considering the availability and representativeness of the data, this study selects emission of the industrial wastewater discharge ($\mathit{water}$), industrial sulfur dioxide emission ($so2$), and industrial soot and dust emission ($\mathit{dust}$) of each city to characterize the environmental pollution to present the impact of the development of the digital economy on various environmental pollutants. In addition, haze, as a major air pollutant, is becoming an important environment problem to be solved in China. Therefore, the annual average PM2.5 concentration ($pm2.5$) of Chinese cities released by the Air Composition Analysis Group of Dalhousie University is included in the explained variables. We take the adoption of four different pollutant indicators adding into regress model separately, which not only aim to comprehensively investigate the impact of digital economy development on various environmental pollutants, but also increase the robustness of research results from mutual confirmation.

2. Explanatory variable. Concerning Zhao et al. (2022), this paper builds the development index of the digital economy ($\mathit{de}$) from two aspects: the Internet development and digital finance. The Internet development level is obtained by synthesizing four aspects indices through principal component analysis, which includes the Internet penetration rate, employees of the Internet, Internet economy scale, and mobile phone penetration (Guo et al. 2020b). Specifically, the Internet penetration rate is expressed by the number of Internet broadband access users, and the mobile phone penetration rate is measured by the number of mobile phone users. Employees of the Internet is calculated by the number of employees engaging in the transmission of computer services and software. And telecom business income is chosen as the proxy variable for the Internet economy scale. The digital finance level is expressed by the digital inclusive finance index jointly prepared by Peking University and Ant Financial Ltd. Based on the principal component analysis, we integrate above five specific indicators into a comprehensive index into characterize the development index of the digital economy.

3. Mediating variables. Referring to Yan et al. (2022) and Xu et al. (2022), we take the ratio of the output value of tertiary industry to secondary industry to represent the industrial structure upgrading index ($\mathit{is}$). According to the previous discussion on the influence mechanism, the related industries about the digital economy mainly belong to the tertiary industry. Therefore, industrial upgrading can be measured by the expansion of the proportion of the tertiary industry in the economic structure. Meanwhile, the industrial structure upgrading index reflects the degree of industrial structure transformation from industrialization to service. The larger the indicator value, the higher the level of service of the industrial structure, and the more pronounced the upgrading of the industrial system.

Drawing on the research by Yao et al. (2022), this paper used the number of green patents applied each year in the city to indicate green technology innovation ($\mathit{gp}$), rather than the patent approval. The main reason for making such a choice is based on the following considerations: First, because it would take some time from patent application to authorization, the number of patent applications reflects the innovation achievements of the enterprise in the current period compared with the number of grants. Second, whether the patent will be granted is not solely affected by the level of technology, in fact, it also affects by patent licensing preference of patent agencies, politics and other aspects (Xu et al. 2021). Obviously, based on the above reasons, the number of green patent applications is a more appropriate indicator for green technological innovation.

• (4)Control variables. To evaluate the pollution reduction effect of digital economy development more accurately and weaken the impact of missing variables on the model, following control variables that may affect environmental pollution emission were added into regression model. The city size ($\mathit{ps}$) is described by the population density (Gu et al. 2022). R&D investment ($\mathit{tc}$) is defined as the percentage of S&T expenditure in local fiscal expenditure (Du et al. 2020). Urbanization ($\mathit{ur}$) is measured by the percentage of urban construction land in the total area of the city. The share of public financial expenditure in GDP is chosen to characterize local government intervention level to local economy ($\mathit{gov}$) (Li et al. 2022b). Foreign direct investment ($\mathit{fdi}$) is calculated by the proportion of actual FDI in GDP (Yi et al. 2022); The transportation infrastructure ($\mathit{inf}$) is expressed in per capita urban road area.

Data sources and descriptive statistics

Considering the availability and reliability of data, this paper selected 280 prefecture-level cities in China from 2011 to 2019 as research samples, covering 30 provinces. Relevant data mainly comes from the China Urban Statistical Yearbook (2012–2020), the China Urban Construction Statistical Yearbook (2012–2020), the State Intellectual Property Office of China and China Statistical Yearbook (2012–2020). Few missing data are supplemented by average growth rate or average value. To reduce the impact of heteroscedasticity, all variables are logarithmized in regression.

Table 1 reports the descriptive statistical results after logarithmization of each variable. It can be found that there are apparent differences in the data of all variables. For example, the maximum value and the minimum value of the digital economic index ($\mathit{lned}$) is 13.488 and 10.370, which indicates that the imbalance of development among different cities in China exist objectively. At the same time, we calculated the variance inflation factor (VIF) of every variable. The results shown that the maximum VIF is 4.60, the minimum VIF is 1.49, and the average VIF is 2.51, which are far less than the critical value of 10, indicating that there is no multicollinearity problem.

Table 1.

Descriptive statistics

Variable	Obs	Mean	SD	Min	Median	Max
$lnpm2.5$	2520	3.722	0.310	3.204	3.706	4.298
$\mathit{lnwater}$	2520	8.187	1.010	6.116	8.266	9.910
$lnso2$	2520	10.040	1.072	7.918	10.110	11.690
$\mathit{lndust}$	2520	9.711	1.018	7.748	9.781	11.387
$\mathit{lnde}$	2520	11.807	0.830	10.370	11.734	13.488
$\mathit{lnis}$	2520	-0.124	0.410	-0.827	-1.154	0.704
$\mathit{lngp}$	2520	5.172	1.536	2.603	5.050	8.172
$\mathit{lned}$	2520	10.727	0.587	8.773	10.693	15.675
$\mathit{lnps}$	2520	5.780	0.798	4.156	5.903	6.907
$\mathit{lntc}$	2520	0.130	0.808	-1.255	0.146	1.571
$\mathit{lnur}$	2520	1.672	0.979	-0.199	1.721	3.330
$\mathit{lngov}$	2520	2.966	0.479	2.272	2.897	4.081
$\mathit{lnfdi}$	2520	-0.123	1.250	-2.926	0.139	1.597
$\mathit{lninf}$	2520	2.430	0.477	1.499	2.463	3.231

Empirical results and analysis

Benchmark regression

According to the results of the Hausman test (p < 0.01), we selected the fixed effect model to conduct empirical regression analysis. The benchmark regression results of the digital economy on four pollutants are listed in Table 2. The result from columns (1) to (4) shows that the impact of the digital economy development on the four pollutants was negative and passed the significance test, indicating that the digital economy development will significantly inhibit the emission of various pollutants. The research hypothesis H1 is supported preliminary. Specifically, for different environmental pollutants, the emission reduction effects of the digital economy development exhibit certain heterogeneity. Where, every 1% increase in the digital economy index will reduce industrial sulfur dioxide emissions by 0.311% (α = -0.311, p < 0.01). It is followed by industrial wastewater emissions (α = -0.236, p < 0.01), industrial soot and dust emissions (α = -0.167, p < 0.01) and PM2.5 concentration (α = -0.113, p < 0.01). Therefore, the rank of the impact of the development of digital economy on the four pollutants can be summarized briefly as follows: industrial sulfur dioxide > industrial wastewater > industrial soot and dust > PM2.5.

Table 2.

Results of benchmark regression

	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)
$\mathit{lnde}$	-0.113***	-0.236***	-0.311***	-0.167***
	(-11.85)	(-8.26)	(-6.63)	(-4.26)
$\mathit{lnps}$	-0.825***	-0.262	-3.478***	-1.206***
	(-9.01)	(-0.95)	(-7.74)	(-3.22)
$\mathit{lntc}$	-0.038***	-0.075***	-0.142***	-0.053
	(-4.67)	(-3.09)	(-3.57)	(-1.60)
$\mathit{lnur}$	0.041***	0.176***	0.283***	0.091**
	(4.51)	(6.54)	(6.41)	(2.48)
$\mathit{lngov}$	-0.132***	-0.494***	-0.868***	-0.585***
	(-13.80)	(-17.32)	(-18.55)	(-15.00)
$\mathit{lnfdi}$	0.038***	0.064***	0.180***	0.062***
	(8.78)	(5.00)	(8.57)	(3.56)
$\mathit{lninf}$	-0.183***	-0.441***	-0.982***	-0.615***
	(-12.84)	(-10.37)	(-14.08)	(-10.58)
$\mathit{constant}$	10.600***	14.748***	38.338***	21.744***
	(19.88)	(9.24)	(14.66)	(9.96)
$R^2$	0.307	0.256	0.325	0.197
Obs	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

Influence mechanism tests

To verify the influence mechanism of the digital economy on pollution reduction effect, this part focuses on the mediating role of industrial structure upgrading and green technology innovation. As shown in Table 3, columns (1) and (2) respectively reports the regression estimation results of the digital economy index ($\mathit{lnde}$) on industrial structure upgrading ($\mathit{lnis}$) and green technology innovation ($\mathit{lngp}$), that is, the empirical test results of Eq. (2). Columns (3) to (10) are the regression estimation results of Eq. 3, where columns (3), (5), (7) and (9) respectively represent the transmission path of the digital economy to achieve emission reduction effect on four pollutants through industrial structure upgrading, and green technology innovation is shown in columns (4), (6), (8) and (10).

Table 3.

Results of mediating effect test

	$\mathit{lnis}$	$\mathit{lngp}$	$lnpm2.5$	$lnpm2.5$	$\mathit{lnwater}$	$\mathit{lnwater}$	$lnso2$	$lnso2$	$\mathit{lndust}$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
$\mathit{lnde}$	0.146***	0.466***	-0.068***	-0.044***	-0.138***	-0.096***	-0.097**	0.0002	-0.054	-0.001
	(9.48)	(12.18)	(-8.06)	(-5.5103)	(-5.07)	(-3.56)	(-2.31)	(0.01)	(-1.42)	(-0.03)
$\mathit{lnis}$			-0.309***		-0.673***		-1.464***		-0.771***
			(-27.22)		(-18.40)		(-25.97)		(-15.07)
$\mathit{lngp}$				-0.149***		-0.299***		-0.667***		-0.355***
				(-35.32)		(-20.67)		(-30.68)		(-17.52)
$\mathit{lnps}$	0.353**	3.998***	-0.716***	-0.227***	-0.024	0.936***	-2.962***	-0.813**	-0.935***	0.214
	(2.37)	(10.90)	(-9.01)	(-3.02)	(-0.09)	(3.63)	(-7.51)	(-2.10)	(-2.61)	(0.59)
$\mathit{lntc}$	-0.030**	0.247***	-0.047***	-0.001	-0.095***	-0.001	-0.185***	0.023	-0.076**	0.035
	(-2.26)	(7.60)	(-6.66)	(-0.15)	(-4.20)	(-0.05)	(-5.30)	(0.68)	(-2.408)	(1.09)
$\mathit{lnur}$	-0.083***	-0.254***	0.015*	0.003	0.120***	0.100***	0.161***	0.113***	0.027	0.001
	(-5.72)	(-7.06)	(1.89)	(0.34)	(4.75)	(4.00)	(4.14)	(3.02)	(0.77)	(0.02)
$\mathit{lngov}$	0.408***	0.827***	-0.005	-0.008	-0.219***	-0.247***	-0.269***	-0.316***	-0.270***	-0.292***
	(26.56)	(21.65)	(-0.54)	(-0.95)	(-7.19)	(-8.56)	(-5.73)	(-7.33)	(-6.34)	(-7.24)
$\mathit{lnfdi}$	-0.061***	-0.163***	0.019***	0.013***	0.023*	0.015	0.091***	0.071***	0.015	0.004
	(-8.82)	(-9.53)	(4.97)	(3.76)	(1.90)	(1.26)	(4.85)	(3.95)	(0.91)	(0.26)
$\mathit{lninf}$	0.308***	0.899***	-0.087***	-0.048***	-0.234***	-0.178***	-0.530***	-0.382***	-0.378***	-0.296***
	(13.44)	(15.79)	(-6.80)	(-4.01)	(-5.67)	(-4.18)	(-8.35)	(-6.20)	(-6.55)	(-5.15)
$\mathit{constant}$	-5.712***	-27.708***	8.833***	6.460***	10.902***	6.451***	29.974***	19.867***	17.341***	11.903***
	(-6.64)	(-12.97)	(18.93)	(14.59)	(7.26)	(4.25)	(12.95)	(8.73)	(8.26)	(5.61)
$R^2$	0.414	0.424	0.479	0.555	0.354	0.376	0.482	0.525	0.271	0.294
$\mathit{Obs}$	2520	2520	2520	2520	2520	2520	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

The results of columns (1) and (2) show that the coefficients of the digital economy index for industrial structure upgrading and green technology innovation are 0.146 and 0.466, respectively, both pass the significance test at the 1% level, which indicates that the development of the digital economy is conducive to promoting the upgrading of industrial structure and the level of green technology innovation. In columns (3), (5), (7) and (9), the coefficients of industrial structure upgrading ($\mathit{lnis}$) for pm2.5 (lnpm2.5), industrial wastewater discharge ($\mathit{lnwater}$), industrial sulfur dioxide emission (lnso2) and industrial soot and dust emission ($\mathit{lndust}$) are significantly negative at 1% level (-0.309; -0.673; -1.464; -0.771), indicating that the development of digital economy reduces the emission of environmental pollutants through industrial structure upgrading. In columns (4), (6), (8) and (10), the coefficients of green technology innovation ($\mathit{lngp}$) on pm2.5 ($lnpm2.5$), industrial wastewater discharged ($\mathit{lnwater}$), industrial sulfur dioxide emission (lnso2) and industrial soot and dust discharge ($\mathit{lndust}$) are negative (-0.149; -0.299; -0.667; -0.355), and passed the significance test at the 1% level, indicating the digital economy has reduced the emission of environmental pollutants by improving the level of green technology innovation. To sum up, Hypothesis 2 has been validated.

According to Fritz and MacKinnon (2007), their research showed that the testing method of the classical mediating effect is too strict to accept the mediating effect. Therefore, to increase the robustness of the influence mechanism test, we introduced Sobel test and non-parametric Bootstrap method to verify further the robustness of the transmission mechanism of industrial structure upgrading and green technology innovation on the pollutants reduction effect of the digital economy, specific results are shown in Table 4. It can be found that, the total effect of the digital economy development on reducing PM2.5 concentration by promoting industrial structure upgrading and improving green technology innovation level is -0.045 and -0.070 respectively, which are significant at 1% level even in Sobel test. Similarly, the total effects of digital economy on industrial wastewater emissions are -0.098 and -0.140 respectively, on industrial sulfur dioxide emissions are -0.214 and -0.311 respectively, and on industrial smoke and dust emissions are -0.1135 and -0.166 respectively, which also pass the significance test at 1% level. For the nonparametric Bootstrap test, the key to the establishment of mediating effect is to observe whether the 95% confidence interval for the deviation correction of indirect effects contains zero. If it does not contain zero, it means the upgrading of industrial structure and the upgrading of industrial structure act as mediating variable indeed. Otherwise, mediating effect is rejected (Preacher and Hayes 2004). According to the Bootstrap test results in Table 4, it can be found that the 95% confidence interval does not contain zero in all empirical result, and the size and significance of the indirect effect are consistent with the Sobe test results, which indicates that the intermediary role of industrial structure upgrading and green technology innovation in the pollution reduction effect of the digital economy has been further verified, and the result of mediating effect is robust.

Table 4.

Results of Sobel and Bootstrap test

Explained variable	Mediating	Sobel test			Bootstrap test
		Sobel value	Indirect effect	Direct effect	Indirect effect	Direct effect	Indirect effect 95% confidence interval
$lnpm2.5$	$\mathit{lnis}$	-0.045*** (-8.95)	-0.045*** (-8.95)	-0.068*** (-8.06)	-0.045*** (-7.30)	-0.068*** (-7.75)	[-0.059, -0.034]
	$\mathit{lngp}$	-0.070*** (-11.52)	-0.070*** (-11.52)	-0.044*** (-5.51)	-0.070*** (-11.28)	-0.044*** (-6.24)	[-0.082, -0.058]
$\mathit{lnwater}$	$\mathit{lnis}$	-0.098*** (-8.42)	-0.098*** (-8.42)	0.138*** (-5.07)	-0.098*** (-7.12)	-0.138*** (-4.51)	[-0.131, -0.076]
	$\mathit{lngp}$	-0.140*** (-10.5)	-0.140*** (-10.50)	-0.096*** (-3.56)	-0.140*** (-9.10)	-0.096*** (-3.46)	[-0.175, -0.116]
$lnso2$	$\mathit{lnis}$	-0.214*** (-8.90)	-0.214*** (-8.90)	-0.097** (-2.31)	0.214*** (-7.15)	-0.094** (-2.20)	[-0.274, 0.155]
	$\mathit{lngp}$	-0.311*** (-11.32)	-0.311*** (-11.32)	0.0002 (0.01)	-0.311*** (-10.12)	0.0002 (0.01)	[-0.377, -0.242]
$\mathit{lndust}$	$\mathit{lnis}$	-0.113*** (-8.02)	-0.113*** (-8.02)	-0.054 (-1.42)	-0.113*** (-7.22)	-0.054 (-1.22)	[-0.142, -0.083]
	$\mathit{lngp}$	-0.166*** (-10.00)	-0.166*** (-10.00)	-0.001 (-0.03)	-0.166*** (-8.93)	-0. 0 (-0.02)	[-0.207, -0.136]

^*** p < 0.01, ** p < 0.05, * p < 0.1; the values in parentheses denotes z values; Bootstrap sampling times is 300

Robustness test

To verify the robustness of the benchmark conclusions, we adopted the following methods to carry out the robustness test.

Instrumental variable method

Although some control variables were added in the benchmark model, it still is unable to totally rule out the problem caused by missing variables (such as government subsidies, urban resource endowment, etc.). Moreover, there may be a reverse causal relationship between the digital economy and environmental pollution. Therefore, we introduced two stage least square (2SLS) to alleviate endogeneity. Referring to Zhang et al. (2022b), this paper selected the spherical distance between each city and the coastal port as the instrumental variable. However, due to the above instrumental variable does not change with time and cannot be directly used in the panel econometric model. Thus, it is necessary to build a panel instrumental variable that changes with time. Specifically, we construct interaction terms between the time variability of the annual digital economic development level of each city and the spherical distance as the instrumental variable, where the time variability of instrumental variables is represented by the number of internet users in the last year during the sample period (Nunn and Qian 2014).

Column (1) in Table 5 shows the regression results in the first stage of 2SLS, indicates that the coefficient of the instrumental variable ($\mathit{iv}$) to the digital economic index ($\mathit{lnde}$) is significantly positive, which preliminarily verifies its explanatory ability as an instrumental variable. Columns (2) to (5) shows the regression results in the second stage. The results of Kleibergen Paap rk LM test significantly reject the original hypothesis that the instrumental variable is not identified enough. The statistic of Kleibergen Paap Wald rk F is significantly greater than the critical value of the Stock Yogo test of 16.38, which means it rejects original hypothesis of the weak identification test of instrumental variables, further proves that instrumental variables are valid. It is shown from the regression results of 2SLS in Tables 2, 3, 4, and 5 that, the coefficients of the digital economy index for the four pollutants are significantly negative at the level of 1%, indicating that after considering endogeneity, the research conclusion that the development of the digital economy can achieve the effect of pollution reduction still can be testified.

Table 5.

Regression estimation results of 2SLS

Variable	The first stage	The second stage
	$\mathit{lnde}$	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)	(5)
$\mathit{lnde}$		-0.108***	-0.224***	-0.284***	-0.155***
		[-11.90]	[-6.93]	[-5.51]	[-3.38]
$\mathit{iv}$	0.866***
	(50.78)
$\mathit{lnps}$	0.030	-0.829***	-0.271	-3.499***	-1.215***
	(0.36)	[-8.387]	[-1.04]	[-7.03]	[-3.32]
$\mathit{lntc}$	0.014***	-0.038***	-0.075***	-0.142***	-0.053
	(2.97)	[-4.725]	[-3.03]	[-3.53]	[-1.60]
$\mathit{lnur}$	-0.005	0.041***	0.177***	0.284***	0.092**
	(-0.11)	[4.11]	[6.27]	[4.97]	[2.43]
$\mathit{lngov}$	0.004	-0.132***	-0.495***	-0.870***	-0.586***
	(0.70)	[-15.90]	[-17.01]	[-20.31]	[-16.16]
$\mathit{lnfdi}$	0.003	0.038***	0.064***	0.180***	0.063***
	(0.73)	[9.64]	[4.96]	[9.90]	[4.06]
$\mathit{lninf}$	0.012	-0.184***	-0.444***	-0.987***	-0.618***
	(1.14)	[-10.86]	[-9.29]	[-11.83]	[-9.04]
city	Yes	Yes	Yes	Yes	Yes
constant	-4.447***	11.719***	15.976***	40.991***	22.614***
	(-7.87)	[18.00]	[9.06]	[12.40]	[9.29]
Kleibergen-Paap rk LM	216.09***	216.092***	216.092***	216.092***	216.092***
Kleibergen-Paap Wald rk F	2578.65	2578.646	2678.646	2678.646	2678.646
	{16.38}	{16.38}	{16.38}	{16.38}	{16.38}
$R^2$		0.843	0.868	0.684	0.757
$\mathit{Obs}$	2520	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.1; the parentheses are t values, the middle brackets are Z values, and the brackets are the critical values at the 10% level of the Stock-Yogo weak identification test

Exogenous shock test

To further identify the causal relationship between the digital economy and pollutants reduction, drawing on Zou and Pan (2022), we took the network infrastructure upgrade of the Broadband China strategy as an exogenous policy shock, and applicated the difference-in-difference (DID) method to evaluate the pollutants reduction effect of the digital economy (Wang et al. 2022b). In 2014, 2015, and 2016, the Chinese government successively announced three batches of 120 cities (groups) as the Broadband China strategy pilot. The pilot city mainly focuses on improving broadband network speed, expanding network coverage, increasing the number of broadband users, to serve economic and social development. With the progress of network infrastructure, the pilot cities become the leading regions of the digital economy development in China, and the local digital economy has made significant progress under the active promotion of policies. Therefore, the policy of the Broadband China strategy provides an excellent quasi-natural experiment for studying the pollution reduction effect of the digital economy.

Since the Chinese government has successively promoted the Broadband China strategy pilot policy in three batches in 2014, 2015 and 2016, the pilot cities are also added to the experimental group in batches. So, we constructed a multi-period DID model as shown in Eq. (4) to verify whether the Broadband China strategy has achieved pollution reduction effect.

$$ln{Y}_{\mathit{it}}={\eta }_0+{\eta }_1{\mathit{did}}_{\mathit{it}}+\eta X_{\mathit{it}}+{\epsilon }_{\mathit{it}}$$ 4

where, $\mathit{did}$ is the multiplication term between the processing variable (${\mathit{treat}}_i$) and the time variable (${\mathit{post}}_{\mathit{it}}$). The processing variable is a dummy variable to distinguish between the experimental group and the control group. If the city $i$ belongs to the Broadband China strategy pilot cities, and value of ${\mathit{treat}}_i$ is 1. On the contrary, it belongs to the control group and value of ${\mathit{treat}}_i$ is 0. The variable ${\mathit{post}}_{\mathit{it}}$ also is a dummy variable. When the sample is located at the years before the policy implement, the value of ${\mathit{post}}_{\mathit{it}}$ is 0. And after the time of policy implement, the value ${\mathit{post}}_{\mathit{it}}$ is 1. The coefficient ${\eta }_1$ of $\mathit{did}$ indicates the emission reduction effect of the Broadband China strategy pilot policy. The other variables are the same as those in Eq. (1).

Table 6 reports the difference-in-difference estimation results of the Broadband China strategy pilot policy on environmental pollutants, shows that the coefficients of the Broadband China strategy pilot policy for PM2.5 concentration ($lnpm2.5$), industrial wastewater discharge ($\mathit{lnwater}$), industrial sulfur dioxide emissions ($lnso2$), and industrial soot and dust emissions ($\mathit{lndust}$) are significantly negative at the level of 1%, indicating that the policy indeed contributes to the reduction of various pollutants and the main research conclusion of benchmark regression is still robust even considering exogenous policy shocks.

Table 6.

Difference-in-difference estimation results

Variable	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)
$\mathit{treattime}$	-0.150***	-0.167***	-0.744***	-0.270***
	(-15.85)	(-5.52)	(-16.13)	(-6.91)
$\mathit{lnps}$	-0.804***	-0.364	-3.093***	-1.179***
	(-8.60)	(-1.22)	(-6.79)	(-3.05)
$\mathit{lntc}$	-0.039***	-0.074***	-0.133***	-0.054
	(-4.64)	(-2.77)	(-3.27)	(-1.56)
$\mathit{lnur}$	0.042***	0.188***	0.279***	0.101***
	(4.61)	(6.43)	(6.27)	(2.67)
$\mathit{lngov}$	-0.092***	-0.450***	-0.664***	-0.459***
	(-9.58)	(-14.63)	(-14.18)	(-11.56)
$\mathit{lnfdi}$	0.031***	0.060***	0.145***	0.038**
	(6.98)	(4.26)	(6.78)	(2.10)
$\mathit{lninf}$	-0.195***	-0.498***	-0.970***	-0.666***
	(-13.57)	(-10.85)	(-13.88)	(-11.23)
$\mathit{constant}$	9.074***	12.556***	31.924***	19.358***
	(16.73)	(7.23)	(12.08)	(8.63)
$R^2$	0.360	0.247	0.398	0.212
$\mathit{obs}$	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

The premise of the difference-in-difference method for policy effectiveness evaluation is to meet the parallel trend hypothesis, that is, if the Broadband China strategy pilot policy was not implemented, the samples of the experimental group and the control group will share similar development trend of environmental pollutants. Therefore, we take the event analysis method to test the parallel trend (Jacobson et al. 1993). Specifically, we take the first batch implementation time of the Broadband China strategy pilot as the base year, by adding the dummy variables before and after the policy implementation into the difference-in-difference model (Eq. 4) and refold regression. Figure 1 plots 95% confidence interval for the regression coefficient of every dummy variable, shows the regression coefficients of the dummy variable before the Broadband China strategy pilot implementation are all around 0 for the four pollutants, which indicates there is no significant difference between the trends of environmental pollutants emissions in pilot cities and non-pilot cities and the model design meets the precondition of parallel trend assumption. Besides, Fig. 1 further demonstrates the dynamic policy effects of the Broadband China strategy. It can be found that after the implementation of the policy in 2014, the policy effect gradually increased for the four environmental pollutants, indicating that the gradual promotion of the Broadband China strategy policy has gradually enhanced the emission reduction effect for various environmental pollutants.

Fig. 1.

Parallel trend test. Note: The horizontal axis in the Fig. 1 represents the time before and after the implementation of the Broadband China strategy, the vertical axis represents the regression coefficient, and the short vertical line is the 95% confidence interval corresponding to the robust standard error at the city level

Other robustness tests

In addition, the following robustness tests were taken to further verify robustness: First, we standardized sub-indicators of the digital economy to eliminate the impact of dimensions, and the coefficient of variation method was selected to re-measure the development level of the digital economy, which is expressed by the variable $\mathit{dee}$. Then, we take $\mathit{dee}$ into the robustness test to replace the digital economy index calculated by the principal component analysis method. The results are shown in the columns (1) to (4) in Table 7, found that the regression coefficients of the digital economy on the four pollutants are negative and passed the significant test at the 1% level. In addition, we take the lag phase of the digital economy index ($L.lnde)$ to replace $\mathit{lnde}$, and reinspect regression. The results are shown in columns (5) to (8) in Table 7, found that the regression coefficients of the digital economy lag phase are significantly negative at the 1% level, which also verified the robustness of the benchmark regression.

Table 7.

Results of other robustness tests

Variable	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
$\mathit{dee}$	-0.077***	-0.154***	-0.341***	-0.176***
	(-29.23)	(-18.06)	(-26.14)	(-14.95)
$L.lnde$					-0.105***	-0.239***	-0.363***	-0.194***
					(-9.70)	(-7.46)	(-6.86)	(-4.43)
$\mathit{lnps}$	-0.432***	0.517*	-1.594***	-0.236	-0.819***	-0.037	-3.511***	-1.381***
	(-5.28)	(1.95)	(-3.95)	(-0.65)	(-7.78)	(-0.12)	(-6.83)	(-3.26)
$\mathit{lntc}$	-0.033***	-0.066***	-0.120***	-0.042	-0.028***	-0.081***	-0.139***	-0.057
	(-4.69)	(-2.87)	(-3.41)	(-1.32)	(-3.05)	(-3.03)	(-3.14)	(-1.56)
$\mathit{lnur}$	0.009	0.113***	0.133***	0.014	0.047***	0.198***	0.350***	0.094**
	(1.08)	(4.36)	(3.36)	(0.39)	(4.39)	(6.29)	(6.75)	(2.19)
$\mathit{lngov}$	-0.046***	-0.324***	-0.473***	-0.382***	-0.109***	-0.452***	-0.797***	-0.561***
	(-5.14)	(-11.19)	(-10.71)	(-9.56)	(-10.71)	(-14.99)	(-16.02)	(-13.69)
$\mathit{lnfdi}$	0.019***	0.026**	0.092***	0.017	0.038***	0.069***	0.195***	0.085***
	(4.89)	(2.14)	(4.89)	(1.00)	(8.05)	(4.94)	(8.47)	(4.51)
$\mathit{lninf}$	-0.067***	-0.212***	-0.431***	-0.332***	-0.178***	-0.446***	-1.007***	-0.622***
	(-5.06)	(-4.95)	(-6.60)	(-5.62)	(-11.07)	(-9.34)	(-12.79)	(-9.59)
$\mathit{constant}$	6.508***	6.498***	21.513***	13.001***	10.374***	13.319***	38.859***	23.008***
	(13.62)	(4.20)	(9.11)	(6.09)	(16.94)	(7.35)	(12.99)	(9.33)
$R^2$	0.467	0.331	0.473	0.264	0.250	0.228	0.301	0.189
$\mathit{obs}$	2520	2520	2520	2520	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

Regional heterogeneity analysis

Due to the different stages of industrialization, the economy development and resource bases in various regions of China, there are apparent differences in the regional distribution of digital economy development level and environmental pollutants emission, and the impact of the digital economy on environmental pollution may show region heterogeneity. Therefore, according to the classification criteria of the National Bureau of Statistics of China, this study divided the research samples into three sub-samples: eastern, central and western regions to explore the regional heterogeneity characteristics of the effect of digital economy on environmental pollution reduction. The regression results are shown in Table 8, indicates that the digital economy has only achieved the emission reduction effect on PM2.5 in the eastern region. In the central and western region, the development of the digital economy contributes to reduce the emissions of pollutants such as PM2.5 and industrial wastewater, industrial sulfur dioxide and industrial soot and dust. The emission reduction effect of the digital economy in the regional distribution is characterized by weak in the east and strong in the west.

Table 8.

The results of heterogeneity test

Variable	Eastern regions				Central regions				Western regions
	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
$\mathit{lnde}$	-0.052***	-0.045	-0.075	0.042	-0.130***	-0.323***	-0.410***	-0.296***	-0.117***	-0.251***	-0.305***	-0.146**
	(-2.92)	(-1.03)	(-0.81)	(0.60)	(-7.73)	(-6.14)	(-4.90)	(-3.86)	(-8.01)	(-4.95)	(-4.55)	(-2.57)
$\mathit{lnps}$	-1.200***	-0.491	-5.182***	-1.215**	-0.135	1.215**	-1.714*	-1.051	-0.679***	-1.124**	-2.086***	-1.358**
	(-8.56)	(-1.44)	(-7.12)	(-2.22)	(-0.74)	(2.12)	(-1.88)	(-1.26)	(-4.38)	(-2.10)	(-2.94)	(-2.25)
$\mathit{lntc}$	-0.020	-0.009	-0.023	0.028	-0.113***	-0.231***	-0.396***	-0.169***	0.024*	0.026	0.063	0.059
	(-1.41)	(-0.27)	(-0.32)	(0.50)	(-8.18)	(-5.35)	(-5.76)	(-2.68)	(1.68)	(0.53)	(0.96)	(1.06)
$\mathit{lnur}$	0.038***	0.212***	0.268***	0.080*	0.066***	0.147**	0.592***	0.246***	-0.011	0.137**	-0.045	-0.070
	(3.09)	(7.17)	(4.23)	(1.68)	(3.45)	(2.44)	(6.18)	(2.81)	(-0.64)	(2.26)	(-0.57)	(-1.02)
$\mathit{lngov}$	-0.131***	-0.377***	-0.812***	-0.518***	-0.117***	-0.610***	-0.882***	-0.656***	-0.118***	-0.384***	-0.741***	-0.481***
	(-8.56)	(-10.09)	(-10.18)	(-8.64)	(-7.71)	(-12.82)	(-11.66)	(-9.47)	(-5.94)	(-5.59)	(-8.16)	(-6.22)
$\mathit{lnfdi}$	0.052***	0.119***	0.252***	0.076***	0.048***	-0.039	0.211***	0.116***	0.018***	0.082***	0.093***	0.032
	(7.13)	(6.65)	(6.60)	(2.64)	(5.21)	(-1.34)	(4.57)	(2.76)	(2.90)	(3.87)	(3.30)	(1.33)
$\mathit{lninf}$	-0.123***	-0.410***	-0.914***	-0.386***	-0.133***	-0.346***	-0.890***	-0.698***	-0.271***	-0.491***	-1.165***	-0.667***
	(-4.64)	(-6.35)	(-6.62)	(-3.73)	(-5.56)	(-4.62)	(-7.47)	(-6.39)	(-11.58)	(-6.06)	(-10.88)	(-7.31)
$\mathit{constant}$	12.411***	13.913***	47.185***	19.065***	6.648***	7.292**	28.326***	22.470***	9.549***	18.784***	29.648***	21.527***
	(14.20)	(6.55)	(10.38)	(5.58)	(6.21)	(2.17)	(5.31)	(4.60)	(11.59)	(6.59)	(7.87)	(6.71)
$R^2$	0.323	0.306	0.334	0.159	0.322	0.326	0.362	0.247	0.393	0.215	0.357	0.208
$\mathit{obs}$	909	909	909	909	873	873	873	873	738	738	738	738

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

Threshold analysis

Based on the previous discussion, it is further worth noting that the validity of the digital economy for environmental pollutants reduction is subject to some constraints. Among various factors, there is no doubt that the level of economic development is the most essential factor affecting environmental pollution. Only the areas with certain economic foundation could provide sound and reliable technical platforms, factor endowments, financial support and data information support system for the development of the digital economy (Baiocchi et al. 2010; Han et al. 2022). At the same time, with the evolution of the external environment, it also determines that the emission reduction effect of the digital economy on environmental pollution will change correspondingly (Li et al. 2021). Therefore, referring to the panel threshold model proposed by Hansen (1999), this paper used the level of economic development as a threshold variable to construct the following panel threshold regression equation:

$$\begin{array}{c}\ln Y_{\mathit{it}}={\eta }_0+{\eta }_1{\mathrm{lnDE}}_{\mathrm{it}}·\mathrm{I}\left(\mathrm{l},n,E,D,\le ,{\gamma }_1\right)+{\eta }_2{\mathrm{lnDE}}_{\mathrm{it}}·\mathrm{I}\left({\gamma }_1,<,\mathrm{l},n,E,D,\le ,{\gamma }_2\right)+\cdots +\\ {\eta }_{\mathrm{n}-1}{\mathrm{lnDE}}_{\mathrm{it}}·\mathrm{I}\left({\gamma }_{\mathrm{n}-1},<,\ln ,E,D,\le ,{\gamma }_{\mathrm{n}}\right)+{\eta }_{\mathrm{n}}{\mathrm{lnDE}}_{\mathrm{it}}·\mathrm{I}\left({\gamma }_{\mathrm{n}},<,\ln ,E,D\right)+\eta X_{\mathrm{it}}+{\epsilon }_{\mathit{it}}\end{array}$$ 5

where $\mathit{ed}$ is the threshold variable, represented by the level of economic development, measured by the urban per capita GDP (Deng et al. 2022). I (•) is an indicator function, γ represents the unknown threshold value, and ${\gamma }_1<{\gamma }_2<···<{\gamma }_n$. The model is similar to a piecewise function model, that is, when the threshold variable value is less than γ₁, the regression coefficient of the explanatory variable to the explained variable is ${\eta }_1$. When the threshold variable value is between γ₁ and γ₂, the regression coefficient of the explanatory variable to the defined variable is ${\eta }_2$, and so on. Other variables meaning is as same as in Eq. (1). Based on this, the Bootstrap method is used to test the existence of the threshold effect of economic development level, and the results are shown in Table 9.

Table 9.

Results of threshold existence test

Threshold variable	Explained variable	Threshold type	F value	P value	BS times	critical value
						10%	5%	1%
$\mathit{ed}$	$lnpm2.5$	Single threshold	190.24***	0.000	300	72.395	80.299	96.795
		Double threshold	86.17***	0.000	300	42.639	46.439	53.592
		Triple threshold	71.01	1.000	300	165.809	178.348	203.321
$\mathit{ed}$	$\mathit{lnwater}$	Single threshold	89.42***	0.000	300	40.686	48.310	57.437
		Double threshold	39.24**	0.047	300	31.978	37.043	46.222
		Triple threshold	44.48	0.830	300	83.826	93.125	110.644
$\mathit{ed}$	$lnso2$	Single threshold	112.88***	0.000	300	67.655	73.142	84.982
		Double threshold	109.19***	0.000	300	50.828	56.043	69.966
		Triple threshold	90.06	0.990	300	207.603	221.260	256.292
$\mathit{ed}$	$\mathit{lndust}$	Single threshold	72.53***	0.000	300	38.699	44.097	54.441
		Double threshold	24.36	0.250	300	31.975	35.227	50.215
		Triple threshold	20.51	0.893	300	55.334	64.897	76.717

^*** p < 0.01, ** p < 0.05, * p < 0.10

The results of F value and P value in Table 9 shows that in terms of PM2.5 concentration ($lnpm2.5$), industrial wastewater discharge ($\mathit{lnWater}$), and industrial sulfur dioxide discharge ($lnso2$), there is double threshold on the level of economic development on the above three pollutants and the F value of the threshold variable of economic development level is significant at the level of 1%, 5% and 1% respectively. For industrial soot and dust emissions ($\mathit{lndust}$), the level of economic development shows a single threshold effect at the 1% significant level. The specific threshold estimates are shown in Table 10.

Table 10.

Results of threshold estimates

Threshold variable	Explained variable	Threshold number	Estimated value	95% confidence level
$\mathit{ed}$	$lnpm2.5$	Double threshold	10.365	[10.335, 10.378]
			11.031	[10.998, 11.049]
	$\mathit{lnwater}$	Double threshold	10.104	[10.073, 10.148]
			10.931	[10.848, 10.942]
	$lnso2$	Double threshold	10.365	[10.349, 10.378]
			11.049	[11.005, 11.069]
	$\mathit{lndust}$	Single threshold	10.349	[10.335, 10.365]

Table 11 reports the economic development level threshold regression results of the pollutants reduction effect of digital economy development. Column (1) shows that there is apparent heterogeneity in the emission reduction effect of the digital economy index ($\mathit{lnde}$) on PM2.5 concentration ($lnpm2.5$) in different economic development levels. Specifically, when the level of economic development ($\mathit{lned}$) is less than 10.365, the regression coefficient of the digital economy index to PM2.5 concentration is -0.081, indicating that the development of the digital economy will significantly inhibit PM2.5. concentration in the first threshold interval. When the level of economic development is between 10.365 to 11.031, the estimated coefficient of the digital economy index on PM2.5 becomes − 0.092, which means that the emission reduction effect of digital economy development on PM2.5 concentration has increased compared with the previous threshold. In terms of the level of economic development is greater than 11.031, the regression coefficient of the digital economy index to PM2.5 concentration further becomes -0.099, indicating that the digital economy has the most apparent emission reduction effect on PM2.5 concentration in the third threshold interval.

Table 11.

The results of regression estimation of threshold model

Threshold interval	$lnpm2.5$	$\mathit{lnwater}$	$lnso2$	$\mathit{lndust}$
	(1)	(2)	(3)	(4)
$\mathit{lned}$≤10.3645	-0.081*** (-8.70)
10.3645 < $\mathit{lned}$≤11.0314	-0.092 (-9.99)***
$\mathit{lned}$>11.0314	-0.099*** (-10.82)
$\mathit{lned}$≤10.1036		-0.152*** (-5.22)
10.1036 < $\mathit{lned}$≤10.9313		-0.179*** (-6.29)
$\mathit{lned}$>10.9313		-0.193*** (-6.82)
$\mathit{lned}$≤10.3645			-0.174*** (-3.78)
10.3645 < $\mathit{lned}$≤11.0486			-0.216*** (-4.76)
$\mathit{lned}$>11.0486			-0.257*** (-5.69)
$\mathit{lned}$≤10.3485				-0.098** (-2.47)
$\mathit{lned}$>10.3485				-0.124*** (-3.19)
$\mathit{lnps}$	-0.617*** (-7.01)	0.181 (0.67)	-2.569*** (-5.90)	-0.924** (-2.49)
$\mathit{lntc}$	-0.016** (-2.06)	-0.025 (-1.02)	-0.037 (-0.94)	-0.030 (-0.91)
$\mathit{lnur}$	0.032*** (3.76)	0.161*** (6.12)	0.250*** (5.90)	0.068* (1.86)
$\mathit{lngov}$	-0.085*** (-8.85)	-0.467*** (-13.88)	-0.645*** (-13.58)	-0.524*** (-13.37)
$\mathit{lnfdi}$	0.034*** (8.43)	0.056*** (4.45)	0.168*** (8.33)	0.055*** (3.20)
$\mathit{lninf}$	-0.147*** (-10.75)	-0.362*** (-8.58)	-0.829*** (-12.25)	-0.550*** (-9.50)
$\mathit{constant}$	8.918*** (17.24)	11.098*** (6.95)	30.985*** (12.11)	19.225*** (8.84)
$R^2$	0.377	0.292	0.381	0.219
$\mathit{Obs}$	2520	2520	2520	2520

^*** p < 0.01, ** p < 0.05, * p < 0.10; the values in parentheses denotes t values

Similarly, columns (2)—(4) reports the threshold effect of the impact of the digital economy on the other three pollutants at the different level of economic development. It is not difficult to find that the emission reduction effect of the digital economy development on other pollutants also shows significant marginal increment with the improvement of economic development level. In general, the pollution reduction effect of the digital economy represents from weak to strong with the improvement of urban economic development level.

Conclusion and policy implication

As a new form of economy, the digital economy not only is conducive to unleash economic growth potential and promote the development of emerging industries, as well as an important means to drive the upgrading of industrial structure and promote the green transformation of economy. Based on the panel data of 280 prefecture-level cities in China from 2011 to 2019, this study used multiple econometric models such as fixed effect, mediating effect and threshold effect to empirically test how the development of digital economy achieves the impact of pollutants reduction. The research conclusions of this study are as follows: First, the development of the digital economy has significantly reduced the PM2.5 concentration, industrial wastewater discharge, industrial sulfur dioxide emission and industrial soot and dust emissions. With a series of robustness tests such as 2SLS, difference-in-difference, and the replacement of explanatory variable measurement methods, it can ensure the positive effect of development of the digital economy on pollution reduction. It is shown in the analysis of regional heterogeneity that the environmental pollution reduction effect brought by the development of the digital economy in the central and western regions is more prominent. Secondly, results of mediating effect test indicate the influence mechanism of the digital economy can reduce environmental pollution emissions mainly rely on promoting the upgrading of industrial structure (structural effect) and upgrading the level of green technology innovation (technical effect). Third, the development of digital economy has a threshold effect on the level of economic development to achieve its pollution reduction effect. Further identification of the threshold effect indicates that the higher the level of economic development, the better in emission reduction effect.

According to the above conclusions, this paper puts forward the following suggestions: (1) The government should build more complete digital infrastructure system, vigorously promote the development of high-end information technologies such as 5G, big data, blockchain, cloud computing, and the Internet of Things, to accelerate the construction of new infrastructure in China. At same time, the government has to increase its investment and support for digital technology, expanded the information technology R&D platform and laid a solid foundation for the digital industry through talent introduction, tax subsidies and other related support policies. (2) Based on digital industrialization, China should promote the penetration of the digital economy into traditional industries, improve production efficiency, make data become the core production factors, and enable digital economy to play the role of pollution reduction by promoting the optimization of industrial structure. Moreover, governments should guide financial resources to enterprises with green R&D motivation with digital finance, improve resource utilization efficiency, and promote pollution reduction. Local governments should gradually adapt to local conditions and times to implement digital economy development policies appropriate to local resource advantages, and find a balance point between digital economy and local industrial development mode. (3) It should be fully aware that only the continuous growth of the economy will better protect the development of the digital economy. Therefore, achieving high-quality economic growth remains the top priority for China. Therefore, we should pay attention to the complementarity between economic growth and digital economic development and promote their coordinated development.

It should be pointed out that our research also has some limitations. First, the development of the digital economy may have spatial spillover effects, which means the development of local digital economy will not only affect the local environmental pollution, but also affect the adjacent areas. Limited by research content, the spatial effect of the digital economy has not been included in this paper. Secondly, this paper only verifies the transmission mechanism of the digital economy to achieve the effect of pollution reduction from the two paths of promoting the upgrading of industrial structure and promoting green technology innovation. Due to the complexity of the digital economy and environmental pollution, there may still be other influence mechanisms to be explored. The inspection and identification of other influence mechanisms (such as the degree of marketization) still need to continue to pay attention in the follow-up research.

Author contribution

Qiuqiu Guo contributed to data curation, formal analysis, writing—original draft, and writing—review and editing. Xiaoyu Ma contributed on the ceptualization, methodology, formal analysis, and supervision. Jingrui Zhao contributed to methodology, and writing—review and editing. All the authors read and approved the manuscript.

Funding

This work was supported by Western China Program of the National Social Science Foundation of China (Grant numbers: 21XRK007), National Natural Science Foundation of China (Grant numbers: 71663050).

Data availability

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Conflicts of interest

The authors declare no conflict of interest.