How Does Industrial Digitalization Affect Enterprise Environmental Performance?

Despite the increasing use of digital technology in industrial production, how industrial digitalization affects the environmental performance of production activities remains unclear. This research contributes to the literature on the relationship between industrial digitalization and enterprise environmental performance by employing a large sample of Chinese manufacturing enterprises. Results indicate that the environmental performance of manufacturing enterprises has been signicantly improved in the process of industrial digital transformation. Structural and technology effects are the inuencing mechanisms. Industrial digitalization reduces the production scale of heavy polluting enterprises and improves product innovation and green total factor productivity, but it has an insignicant effect on total factor productivity. Moreover, industrial digitalization improves enterprise environmental performance by introducing front-end cleaner production technologies, rather than by increasing pipe-end pollutant treatment facilities.

However, most studies argue that enterprises not only use ICT to improve process e ciency but also improve the capabilities of product design and service innovation (Neuhofer et al., 2015;Martínez-Caro, 2020). An explanation of ICT Productivity Paradox is that ICT is used to achieve some social goals, such as reducing labor fatigue and reducing pollution, which cannot be observed in traditional statistical indicators. As China has experienced over the past few decades, ICT may help decouple industrial sector growth from various environmental indicators.
The net impact of ICT on the environment is not only mixed in empirical evidence but also theoretically ambiguous (Dedrick, 2010;Lange et al., 2020). Empirically, some studies show the environmental bene ts from ICT adoption (Lu, 2018;Khan, 2019), whereas others nd that ICT adoption increases energy consumption and pollution (Park et al., 2018;Avom et al., 2020). Berkhout and Hertin (2001), Beier et al. (2018), and Kunkel and Matthess (2020) propose a theoretical framework, which classi es the environmental effects of ICT adoption into two aspects, to reconcile the two pieces of con icting evidences. The rst is the direct effect, which means that ICT increases the energy consumption and resource use in the life cycle, thus reducing environmental performance (Berkhout and Hertin, 2001). The second is indirect effect, which indicates that ICT adoption affects the production scale, product structure, and process e ciency, thus affecting environmental performance (Beier et al., 2018;Kunkel and Matthess, 2020). Both effects form a nonlinear relationship between ICT penetration and pollutant emissions, leading to inconsistent conclusions drawn from different samples (Higón, 2017;Avom et al., 2020). Therefore, clarifying how industrial digitalization affects the environment is more important than identifying their correlation.
Enterprise investment in improving environmental performance is the driving force for sustainable economic development, although the government also plays an important role. Exploring the driving factors of enterprise environmental performance is also an important topic in the eld of environmental economics, as the number of studies on the environmental behavior of microenterprises is increasing (Zhang et al., 2020; Wen and Lee, 2020).
However, literature on the in uence and mechanism of how industrial digitalization affects the enterprise environmental performance is limited. Environmental technologies have two types: front-end cleaner production technologies and pipe-end treatment technologies. The type of environmental technology adopted by manufacturing enterprises to achieve their environmental goals has different policy implications to the sustainable economic development. However, identifying which environmental technology is being used on the basis of macro data is di cult. To decouple industrial economic development from environmental indicators in the era of digital economy, further exploring the relationship between industrial digital transformation and the choice of environmental technology is necessary.
This study aims to explore the impact and mechanism of industrial digitalization on the environmental performance of manufacturing enterprises in China. Data about environmental information at the enterprise level are scarce, and most related studies use the data of listed enterprises or survey data from small samples (Hu et al., 2019). Different from most studies, this research uses a unique dataset at the enterprise level from 2002 to 2012, including pollutant information and nancial information. The large sample of microenterprise data leads to some additional interesting ndings in this study, which enriches the literature in this eld of enterprise environmental behavior and the theory of digital economy. The contribution of this research to literature mainly includes the following aspects.
First, it provides a novel explanation for the decoupling of industrial sector growth from various environmental indicators from the perspective of industrial digital transformation. The research not only provides a new perspective for understanding the improvement of environmental performance in China during the process of industrialization but also gives ideas for developing or industrializing countries to explore the pathway of sustainable industrial development. Second, this study enriches the literature on ICT Productivity Paradox.
Although industrial digitalization does not improve productivity, it enhances the green total factor productivity (GTFP) and environmental performance. ICT investment or industrial digitalization has brought about many welfare improvements that cannot be observed in traditional statistical indicators. Third, this work identi es the structural and technology effects of industrial digital transformation on the enterprise environmental performance.
Speci cally, it provides a detailed discussion of the in uence of industrial digitalization on the environmental technology choice of enterprises. This study shows that industrial digitalization promotes manufacturing enterprises to adopt front-end cleaner production technologies, rather than pipe-end pollutant treatment technologies. It also suggests that digital transformation is an important driving force for the decoupling of industrial development and environmental indicators in the era of digital economy.
The remainder of this paper is structured as follows: Section 2 provides a review of the literature and hypotheses. Section 3 introduces the data, variables, and econometric models. Section 4 presents the empirical results. Section 5 investigates the role of technology factors in the environmental effects of industrial digital transformation. The nal section concludes.

Literature review
Industrialization is one of the most important determinants of changes in pollution emissions, and the mantra of scholars and media is that industrial economic development is inseparable from environmental pollution (Cherniwchan, 2012). Therefore, exploring the driving forces of decoupling industrial economic development from environmental indicators is always an important topic in the eld of environmental economics (Hu et al., 2020). Research on the economic effects of ICT can be traced back to production theory. Ever since Robert Solow proposed the Productivity Paradox, many studies have been conducted on how ICT affects productivity. Although uncertainty remains, the increasing application of ICT in the industrial sector has triggered great hopes of improving productivity and reducing pollution emissions (Higón, 2017). ICT penetration helps manufacturing enterprises improve process e ciency, provide better service to customers than before, optimize work practices, and enhance product design (Neuhofer et al., 2015;Martínez-Caro, 2020). Another study suggests that ICT can be used to achieve other goals, resulting in an irrelevance between ICT and productivity. As revealed by DeStefano et al. (2018), manufacturing enterprises adopt ICT to achieve the goal of sales growth, rather than productivity.
The literature on the environmental effects of ICT is mainly based on the direct effects that ICT increases energy consumption and carbon emissions. It indicates that industrial digitalization leads to more energy consumption and poorer environmental performance than usual. Salahuddin  (2019) show that the ICT sector contributes a large amount of carbon emissions due to its energy consumption and intermediate inputs of energy-intensive products. However, some studies suggest that industrial digitalization also affects energy consumption and environmental quality through indirect effects. The indirect effects of digital transformation result from its in uence on factors such as production e ciency, technology progress, and production scale (Kunkel and Matthess, 2020). Industrial digitalization may boost sustainability if the indirect effect is greater than the direct effect (Lange et al., 2020).
The indirect impacts of ICT on environment can be divided into scale effect, structure effect, and technology effect; therefore, the comprehensive effect is uncertain (Hao et al., 2020;Avom et al., 2020). The scale effect refers to the fact that digital transformation promotes industrial expansion and leads to increased pollution. Structural effect indicates that industrial transformation leads to the advancement and rationalization of structure or the reduction of pollution-intensive production activities. Technology effect means that ICT increases productivity and thus leads to improved environmental performance. Lange et al. (2020) discuss the direct effect and three indirect effects in detail and explain that the environmental effects of industrial digitalization depend on the net effect of these four effects.

Research hypothesis
The studies discussed in the literature review section are all macroscopic research, mainly focusing on energy consumption, total environmental pollution, and carbon emissions. These four effects of industrial digitalization also exist in microenterprises, and environmental performance is affected by positive and negative impacts.
However, the research design of the present study mainly focuses on structural and technology effects, suggesting that industrial digitalization has a positive effect on enterprise environmental performance.
The pollutant used in this study is chemical oxygen demand (COD) pollution or water pollution to investigate the choice of environmental technology in manufacturing enterprises. Although many types of pollutants exist, COD is a commonly used pollutant in literature and is mainly determined by the endogenous technology choice of enterprises. Therefore, COD is an ideal indicator of enterprise environmental performance. The direct effects are mainly energy-related pollutants, which are not discussed in this study. That is, the research mainly focuses on the indirect effects of industrial digitalization. Considering that the objective of this study is enterprise environmental performance, scale effects are also controlled in this work. After controlling the output scale factor of enterprises, we nd that industrial digitalization has a positive impact on enterprise environmental performance through technology and structural effects. Hence, this study proposes the following hypothesis: Hypothesis 1.Industrial digitalization or ICT penetration has a signi cant positive impact on the environmental performance of manufacturing enterprises.
Production activities related to pollutants in the industrial sector can be divided into two sub-stages: the production stage and the treatment stage. At the production stage, enterprises produce undesired pollutants during production. Enterprises need the treatment stage to remove undesired pollutants. Two pathways are available to reduce pollutant emissions and improve environmental performance in industrial production activities: either by reducing the volume of produced pollutants or by increasing the volume of removal pollutants (Wang et al., 2021).
Correspondingly, environmental technologies in these two stages are classi ed as cleaner production technologies and pipe-end treatment technologies. Both affect enterprise environmental performance in different ways and have completely different policy implications for industrial sustainable development. Wang et al. (2021) argue that cleaner production technologies have become the dominant approach for pollutant reduction in China. Industrial digital transformation can help enterprises realize the leapfrog transformation of production activities and achieve the goal of improving environmental performance. With the in uence of industrial digital transformation, industrial enterprises can break through the emission reduction pathway of pollution rst and treatment later. Therefore, they tend to adopt cleaner production technologies, rather than increase pipe-end treatment facilities. Hence, this study proposes the following hypothesis: Hypothesis 2.Industrial digitalization signi cantly affects the choice of environmental technologies, and enterprises tend to adopt front-end cleaner production technologies.

Data collection
To verify the proposed hypotheses, this study utilizes the data of a large sample of Chinese manufacturing enterprises from two microenterprise databases. The rst is China Industrial Enterprise Database, which covers all state-owned and non-state-owned industrial enterprises whose main business income is above the designated amount. The second is the Enterprise Pollution Database from the Ministry of Ecology and Environment (or formerly the State Environmental Protection Administration) of the People's Republic of China. The matching of the two micro databases is performed by a team with Beijing Forecast Information Technology Co., Ltd., the company that operates the EPS Database. This study also uses some macro variables from the EPS Database.
The Enterprise Pollution Database, also named the Enterprise Green Development Database (1998-2012), is the most authoritative database for investigating the enterprise environmental performance in China. It mainly reports the information on the production and discharge of water pollutants, gas pollutants, and solid pollutants, including COD and sulfur dioxide (SO 2 ). COD and SO 2 are representative water pollutants and air pollutants, and their discharge is also the proxy indicator of environmental performance commonly used in literature (Clarkson et al., 2011; Wen and Lee, 2020). SO 2 is closely related to energy consumption and is mainly affected by the exogenous intervention of energy policies, whereas COD is determined by the endogenous decision of the production technology. Hence, selecting the production and discharge of COD as the proxy indicators of enterprise environmental performance in this study is reasonable.

Variables de nitions
The dependent variable is enterprise environmental performance, and the pollution emission intensity is selected as the proxy variable in this study. The lower the pollution intensity, the better the enterprise environmental performance. The primary proxy variable, COD Intensity, is de ned as the ratio of COD emissions to the total output of a rm, multiplied by 100 to facilitate the presentation of coe cient. SO 2 Intensity, Sewage Intensity, and some other variables related to COD are also used in this research as dependent variables.
The core explanatory variable is the degree of industrial digital transformation, which is measured by the extent to which ICT is used in industrial production and operations. This study considers two proxy variables. The rst proxy variable, ICT_Capital, is measured by the ratio of ICT capital to industrial added value. The second proxy variable, ICT_Service, is measured by the ratio of ICT service to intermediate inputs in the industry. ICT capital and ICT services are calculated using the input-output table. The ICT data used in the calculation are the ICT investment and ICT services at the provincial-city level, and the kernel density distribution is illustrated in Figure A in the appendix.
The control variables include enterprise characteristics, regional characteristics, and industry characteristics. To analyze the in uence mechanism of industrial digitalization on environmental performance from the perspective of technology factors, this study also introduces some variables, including Product_Inno, total factor productivity (TFP), GTFP, and Technology_Up. The variables of TFP and GTFP are estimated using the Solow residual method. This study also winsorizes the continuous variable at the 0.5 and 99.5 percentiles to leave out extreme outliers. Table 1 provides a detailed description for the de nitions of the variables covered in this study. Table A in the appendix shows the descriptive statistics of the variables. where i, j, p, and t are subscripts referring to the rm, two-digit industry, province, and year, respectively. Pollution is the proxy variable of enterprise environment performance. Digitalization refers to the indicator of digital transformation at the provincial-industry level and is the core explanatory variable of this study. If is signi cant and negative, then it indicates that industrial digital transformation has a positive effect on rm environmental performance. X represents the control variables for enterprise characteristics, Z 1 refers to the control variables for regional characteristics, and Z 2 refers to the control variables for industry characteristics. This study also includes province xed effects μ p and year xed effects λ t to account for the time-invariant regional characteristics and the temporal characteristics of macro-environmental policies, which may affect rm environmental performance, respectively. In the empirical analysis, some regressions also introduce the xed effects of two-digit industries.

Impact of industrial digitalization on enterprise pollution intensity
Manufacturers in the same industry always have the same cleaner production technologies and alternative pollutant treatment technologies at their disposal, and their exposure to industrial technology shocks or other random shocks may be related. Hence, this study employs the robust standard errors adjusted for clustering at the four-digit industry to overcome the cross-sectional correlation among random disturbance terms. Table 2 presents the ndings for the impact of industrial digitalization on enterprise environmental performance. All columns in the table use COD intensity as the dependent variable. Columns (1)-(3) in Table 2 measure the degree of industrial digitalization by the ratio of ICT capital to industrial added value, and Columns (4)-(6) use the intermediate input of ICT service as the measure index. This study considers not only the control variables of the rm characteristics but also those of regional and industry characteristics.
The benchmark results in Table 2 show that industrial digitalization signi cantly improves the environmental performance of manufacturing enterprises. In the table, the coe cients of ICT_Capital are all signi cantly negative at the 5% level, indicating that industrial ICT capital signi cantly reduces the pollution intensity of enterprises. The coe cient of ICT_Service in Column (5)   Notes: The cluster-robust standard errors are shown in brackets. Asterisks ***(1%), **(5%), and *(10%) indicate signi cance at the corresponding levels. Columns (2) and (5) use industry xed effects to control industry characteristics; Columns (3), (6), and (7) control industry characteristic variables, such as trade openness, market competition, and pro t margins.
The regression coe cients of the control variables are basically consistent with the theoretical expectation, indicating that the empirical results are relatively robust and reliable. In terms of rm characteristics, operating scale, foreign direct investment, and state-owned ownership are signi cantly and negatively correlated with COD intensity. Large-scale enterprises can bene t from the advantages of economies of scale and improve their environmental performance. Meanwhile, state-owned enterprises have a strong incentive to pursue socially responsible goals, including environmental quality improvement. Foreign direct investment is also conducive to the improvement of the production technology and productivity level of enterprises. However, a signi cant positive correlation exists between the capital-labor ratio and rm environmental performance. In terms of regional characteristics, the economic growth target constraint of local governments can increase the COD intensity of enterprises; a surrogate relationship is also observed between economic growth and environmental quality to a certain extent. After controlling for the provincial xed effects and the industry characteristics, environmental regulation and technology innovation can reduce the pollution emission intensity and improve the environmental performance of enterprises. Industry characteristics also have signi cant impacts on pollution emission intensity, which is neither discussed in detail here and nor is reported in the table. Notes: The cluster-robust standard errors are shown in brackets. Asterisks ***(1%), **(5%), and *(10%) indicate signi cance at the corresponding levels. This table uses a series of industry-level explanatory variables to control industry characteristics, rather than industry xed effects.
To ensure the robustness of the estimated results, this study conducts a series of robust regression analyses, and the results are shown in Table 3. First, it considers potential threats from the differences in data quality over the years and changes the sample period. Column (1) (4) and (5). Third, sewage intensity is used as the proxy variable of enterprise environmental performance, as presented in Columns (6) and (7). In Table 3, all the coe cients of ICT_Service are signi cantly negative at the 1% level, indicating that the intermediate input of ICT services signi cantly improves the environmental performance of enterprises. ICT_Capital has a signi cant negative impact on COD discharge intensity and sewage discharge intensity, but its effect on SO 2 discharge intensity is insigni cant. The results in Columns (4) and (5) are not contradictory with other empirical results because SO 2 emissions are mainly affected by exogenous factors, such as national energy policies, rather than an endogenous choice of production technology. The empirical results of the robust analysis in Table 3 indicate that the results of this study are robust, and the environmental performance of manufacturing enterprises has been signi cantly improved in the process of industrial digital transformation.

Total emissions effect, structure effect, and network effect
To further understand the impact of industrial digitalization on enterprise environmental performance, this part examines the effects of industrial digitalization in three aspects. First, we use the logarithm of COD emission as the explained variable to investigate the impact of industrial digitalization on the total pollution emissions of enterprises, namely, total emission effect. Second, we examine the structural effect of industrial digitalization, that is, whether industrial digitization limits the production scale of heavy polluters. Third, we examine the network effect of ICT capital and ICT services. That is, as the total amount of ICT capital and ICT services increases, does the environmental effect increase? The empirical test results of total emissions effect and structure effect are shown in Table 4, and those of network effect are presented in Table 5. The empirical results suggest that industrial digitalization not only reduces the COD intensity of enterprises but also reduces the total COD emission, as shown in Columns (1)-(3) in Table 4. Although the coe cient of ICT_Capital in Column (3) is insigni cant, the T value is 1.607. In addition, the coe cients of ICT_Service are signi cantly negative at the 1% level. The COD intensity mentioned above measures the COD emission per unit of output. The variables of ICT_Capital and ICT_Service also have signi cant positive effects on the total output of enterprises; the coe cients of both variables are signi cantly positive, as presented in Columns (5)- (6). Given that ICT technology can promote the total production of enterprises to increase, the negative effects of ICT_Capital and ICT_Service on COD emission indicates that industrial digitalization has a strong impact on the improvement of enterprise environmental performance.
This study introduces the interaction term between COD intensity and these two proxy variables of industrial digitalization in Columns (4)-(6) in Table 4. All the coe cients of ICT_Capital×COD Intensity and ICT_Service×COD Intensity are signi cantly negative at the 1% level, indicating that both proxy variables have signi cant negative impacts on the total output of enterprises. With the increase of ICT application in the industry, manufacturing enterprises with high pollution intensity can reduce their total production scale, and the structure effect of industrial digitalization is established. On the one hand, ICT application increases production exibility and operational agility (Škare and Soriano, 2020); then, manufacturing enterprises can adjust the production plan according to the change of market demand for environmentally friendly products. On the other hand, ICT technology reduces the transaction cost and improves the investment e ciency, which leads to the reduction of the production scale of high-polluting enterprises. Notes: The cluster-robust standard errors are shown in brackets. Asterisks***(1%), **(5%), and *(10%) indicate signi cance at the corresponding levels. This table uses a series of industry-level explanatory variables to control industry characteristics.
In the literature of economic growth theory, general technology has a strong network effect and can exert its positive effect when applied to a large scale. ICT is a typical general technology and has the spillover of network scale. In Table 5, this study introduces the logarithm of total ICT capital and total ICT service input at the regionindustry level as the explanatory variables, which are expressed as ICT_Capital_Network and ICT_Service_Network, respectively. As presented in Table 5, ICT capital has a signi cant negative in uence on the COD emission intensity and the total COD emission at the 5% level, whereas the coe cient of ICT_Service_Network is insigni cant, and the coe cient in Column (6) is positive. These pieces of evidence suggest that the network externalities of ICT capital are in place, but ICT services are not. In this study, samples up to 2012 are used. Digital services were relatively immature during this period; therefore, the ICT services in the manufacturing sector did not exhibit network effect.
The empirical results indicate that not only does the degree of industrial digitalization have a signi cant impact on the improvement of enterprise environmental performance but also the digital transformation can further release the dividend or network effect of the digital economy when it reaches a certain scale.

Industry heterogeneity of environmental effects
Existing studies nd signi cant differences in the degrees of digitalization and pollution intensity among enterprises in different industries. Therefore, this research conducts the following industry heterogeneity analysis of the environmental effects of industrial digitalization. As shown in Table 6, the industries are divided into several categories according to the capital intensity, energy intensity, and COD pollution intensity of the industry.
Although industrial digitalization entirely improves the environmental performance of manufacturing enterprises,   Table 7 shows the empirical results of how industrial digitalization affects the technology progress of enterprises. Speci cally, we employ product innovation (Product_Inno), total factor productivity (TFP), and Green total factor productivity (GTFP) as the proxy variables for technology progress.
The empirical results in Table 7 generally support the view that industrial digitalization promotes enterprise technology progress, but different and interesting results are also obtained. From the empirical results, ICT capital and ICT services have signi cant positive impacts on rm product innovation. As for productivity, the impact of ICT inputs is complex. After controlling for industry characteristics and province xed effects, the coe cients of ICT_Capital and ICT_Service are insigni cant, and this result supports the Solow Productivity Paradox, ICT penetration has not improved the traditional indicator of productivity. When GTFP is used as the explained variable, the coe cient of ICT_Service is signi cantly positive at the 1% level. Meanwhile, the coe cients of ICT_Capital are all positive, and the coe cient of ICT_Capital in Column (5) is signi cant at the 1% level. Therefore, ICT has a positive impact on GTFP. Although ICT does not improve productivity, it has improved GTFP. That is, industrial digitalization has brought about many welfare improvements that cannot be observed in traditional statistical indicators.   (5) and (6), both of which are estimated using the Probit model.

Conclusion And Implication
Driven by the next-generation ICT, digital technology is being embedded in the production of products and services with unprecedented breadth and depth. Based on the actual observation of industrial digital transformation, this study uses intermediate inputs of ICT capital and ICT services to measure industrial digitalization and investigates the impact and mechanism of industrial digitalization on enterprise environmental performance. Using a massive sample of manufacturing enterprises in the period from 2002 to 2012, this study leads to the following main ndings.
In the process of industrial digital transformation, manufacturing enterprises have signi cantly reduced their COD emission intensity and even signi cantly reduced the total COD emission. Combined with the robustness analysis, we conclude that industrial digitalization has a signi cant positive impact on enterprise environmental performance. The environmental effects of industrial digitalization also show signi cant industry heterogeneity.
Industrial digitalization has a great impact on the COD emission intensity of enterprises in heavy polluting industries and non-capital-intensive industries. Empirical evidence suggests that the network effect of ICT capital is in place, whereas ICT services are not. The digital transformation can further release the dividend or network effect of the digital economy when it reaches a certain scale.
In terms of the in uence mechanism, industrial digitalization has two mechanisms on the improvement of enterprise environmental performance: structure effect and technology effect. With the increase of ICT capital and ICT services in the industry, manufacturing enterprises with high pollution intensity can reduce their total production scale; therefore, the hypothesis of structural effect holds. This study employs a series of econometric models to identify the role of technology factors in the relationship between industrial digitalization and enterprise environmental performance. Industrial digitalization has signi cantly increased product innovation and GTFP, but it has an insigni cant effect on TFP. In the process of industrial digital transformation, manufacturing enterprises have improved the environmental performance by introducing front-end cleaner production technologies, rather than increasing pipe-end pollutant treatment facilities. Our ndings also provide an explanation for the Solow Productivity Paradox, and ICT technology has led to social welfare improvements in the environment, rather than traditional productivity indicators.
Our ndings imply that industrial digital transformation plays an important role in sustainable development. ICT has brought about the upgrading of production technology in the manufacturing sector, reducing the amount of pollutants produced in the front-end production process. Promoting the deep integration of digital economy and real economy is an important breakthrough to resolve the contradiction between economic growth and environmental quality, and it is an important driving force to promote sustainable economic development. Our Declarations -Ethical Approval: This is an original article that did not use other information that requires ethical approval.
-Consent to Participate: All authors participated in this article.
-Consent to Publish: All authors have given consent to the publication of this article.
-Authors' Contributions: All authors provided critical feedback and helped shape the research, analysis, and manuscript.
- -Competing Interests: The authors declare that they have no competing interests.