3.2.1 The mechanism of production efficiency.

On the one hand, ICT products have the advantages of knowledge-intensive and efficient circulation and the use of ICT products can optimize the types of enterprise factor inputs and improve the efficiency of resource allocation and production efficiency. Specifically, first of all, ICT products have a substitution effect on traditional production factors such as non-ICT capital and labor. After the enterprise realizes the network and digital transformation through ICT investment, it will change the production mode, organization mode and business practice of the enterprise, and then improve the efficiency of resource allocation and production efficiency. For example, enterprises will realize intelligent production, realize flat enterprise management mode and carry out e-commerce. Second, ICT products have a synergistic effect on traditional production factors. Enterprises will generate data elements when using ICT products. And the industrial Internet platform can collect and process data in a unified way and transmit effective information to different subjects of enterprise production and enterprises related to the upstream and downstream of the industrial chain. These will reduce the degree of information asymmetry and help shorten the time spent on the convergence of factors such as capital and labor. For example, the Internet of Things and big data technologies enable enterprises to monitor production progress in real time, rationally allocate organizational resources and flexibly formulate production plans, thereby improving production efficiency [38]. On the other hand, the improvement of production efficiency will in turn promote green development of enterprises. The negative environmental impact is often related to the excessive consumption of resources and uncontrolled emission of pollutants. The improvement of production efficiency can not only improve the economic benefits of enterprises, but also reduce the redundancy of production input and control the excessive emission of pollutants, which will help promote the green development of enterprises [39]. In summary, this paper proposes hypothesis 2:

1. Hypothesis 2: Production efficiency is one mechanism through which ICT investment impacts on green development of enterprises in highly polluting industries.

3.2.2 The mechanism of green technology innovation.

On the one hand, ICT products have the advantages of externality and spillover effect, and ICT investment can enhance the ability of enterprises to innovate green technology. Specifically, first of all, ICT products accelerate the spillover of knowledge and improve the ability of enterprises to absorb knowledge. Nowadays, a large amount of patent knowledge, green innovation experience and industry dynamics can be searched or inquired through professional innovation knowledge platforms. And this kind of patent knowledge and innovation experience increases rapidly over time. Second, the network platform formed by ICT investment helps to achieve knowledge sharing and green technology R&D cooperation. Enterprises can use the internal network platform to strengthen internal personnel communication and cooperation, and to improve the efficiency of enterprise green technology innovation. Enterprises can also use the external network platform to strengthen cooperation with other enterprises in its supply chain, universities at home and abroad, and scientific research institutions, and to jointly develop more environmentally friendly products. Finally, enterprises can obtain consumers’ consumption and experience data in green products on the online sales platform, analyze consumers’ preferences and suggestions, and then improve the success rate of green product development [40]. On the other hand, green technology innovation also helps enterprises to achieve green development. Green technology innovation can help enterprises to improve production technology and develop more environmentally friendly products in the life cycle [41], which are good for reducing corporate pollution emissions. In summary, this paper proposes hypothesis 3:

1. Hypothesis 3: Green technological innovation is one mechanism through which ICT investment impacts on green development of enterprises in highly polluting industries.

4.1.1 Benchmark regression model construction.

Based on the network effect theory, the impact of ICT investment on green development of enterprises in high-pollution industries is likely to have a threshold effect. At the same time, considering that green development level of the previous period of the enterprise may affect green development level of the current period, this paper chooses the dynamic panel threshold model for research [42,43]. The benchmark regression model is as follows: (1)

In Model (1), subscript i refers to the enterprise, t refers to the year, j refers to the industry, and p refers to the province. Adding ‘L.’ before the variable indicates a lag of one period. LnCEI represents green development level of the enterprise. LnICT represents the level of ICT investment of enterprises, which is both an independent variable and a threshold variable. I(·) represents the indicative function, and when the conditions in the brackets are satisfied, it takes 1, and when it is not satisfied, it takes 0. γ is the threshold value and X represents a series of measurable control variables. δ denotes the fixed effect of the enterprise, λ denotes the fixed effect of year, and ε denotes the random disturbance term.

4.1.2 Mechanism test model construction.

This paper refers to the common practice in economic research, and verifies the existence of the influence mechanism by testing the influence of independent variable on the mechanism variables [44,45]. Because we only want to verify whether ICT investment can affect the mechanism variable, this paper finally chooses the two-way fixed effect model to test. The test model is shown in Model (2). The independent variable in Model (2) occurs before the mechanism variable, so it is relatively exogenous. Therefore, the endogenous problem of the regression is small.

(2)

MNV in Model (2) represent mechanism variables, and the meaning of variable subscripts and other variables in Model (2) are the same as those in Model (1).

4.1.3 Multi-period difference-in-difference (DID) model construction.

The model is used for endogenous test. In 2012, 2013 and 2014, the Chinese government announced three batches of “Smart City” pilot list, which aims to improve the level of urban ICT infrastructure investment. Enterprises can use the ICT infrastructure in the “Smart City” to reduce their carbon emission intensity. Therefore, this paper regards this policy as a quasi-natural experiment and discusses the impact of the policy on the carbon emission intensity of enterprises. Model (3) is constructed for the regression. Model (4) is constructed for parallel trend test of DID model.

(3)

(4)

In Model (3), PilotSC is a dummy variable. If the city c where the enterprise i is located is a “Smart City” in year t, PilotSC takes 1, otherwise 0. In Model (4), PilotSC2 represents the constructed policy occurrence period dummy variable, and the value of m is all integers from -6 to 6. Earlier or later years are merged into the 6th year before or after the actual policy period. In order to avoid the problem of multicollinearity, PilotSC2_i,c,t-1 is deleted in the actual regression.

4.1.4 Sample selection.

Since China’s new accounting standards began to be implemented in 2007, and the latest data on the level of green development of enterprises are only updated to 2019, this paper finally uses the sample of listed companies in the high-pollution industries of China’s Shanghai and Shenzhen A-shares from 2007 to 2019 for research. Enterprises in high-pollution industries are paying more and more attention to the important role of ICT investment in promoting green development. Compared with enterprises in other industries, their carbon emissions are greater, so their potential to use ICT investment to reduce carbon emission intensity will be greater [11]. In 2019, China’s high-polluting industries emitted 8.07 billion tons of carbon dioxide, accounting for 82.4% of total industrial emissions. The delineation of high-pollution industries is based on the standards in “The Plan on the First National Census on Pollution Sources” published by China in 2007. Table 1 shows 11 highly polluting industries, including 5 heavy industries and 6 light industries. In terms of sample processing, samples that were ST, ST* and PT during the sample period were deleted. The samples with total enterprise assets, number of employees and ICT investment less than or equal to 0 during the sample period were deleted. Due to the serious lack of data on economic indicators in Tibet, the samples of listed companies in Tibet were excluded.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. High pollution industry list.

https://doi.org/10.1371/journal.pone.0316337.t001

4.2.1 Dependent variable.

Considering that carbon neutralization is the main goal of current environmental governance in China and carbon emission intensity can reflect the dual connotation of “green” and “development”, this paper finally uses the logarithm of carbon emission intensity (LnCEI) to measure green development level of enterprises [46,47]. Carbon emission intensity (CEI) = carbon emissions of enterprises / operating income of enterprises. The larger the LnCEI, the lower the level of green development of enterprises. Referring to the idea of Chen, this paper uses the carbon emissions at the industry level to calculate the carbon emissions of enterprises [46]. Carbon emissions of enterprises = (carbon emissions of the industry in which the enterprise is located × the operating cost of the enterprise / the main business cost of the industry in which the enterprise is located).

4.2.2 Independent variable.

The data of enterprise ICT investment in existing literature are mostly obtained through sample surveys, leading to a short time span of the research sample [48]. In order to obtain more data for consecutive years, drawing on the idea of Li et al. and Ye et al., this paper use the logarithm of the sum of the capital stock of the three types of ICT products (LnICT), namely, computer equipment, communication equipment and software to measure the level of enterprise ICT investment [11,14]. This data is obtained by manually collating the data in the appendix of the financial statements of listed companies in the RESSET database.

4.2.3 Mechanism variables.

1. Production efficiency. TFP represents the ability of an enterprise to maximize output without increasing resource input. Therefore, this paper uses the logarithm of enterprise TFP (LnTFP) to measure the production efficiency of and choose the Levinsohn-Petrin (LP) method to estimate TFP, which can well reduce endogeneity and sample loss [49].
2. Green technology innovation level. The number of green invention patents shows the substantive green technology innovation level of enterprises [50]. Because the number of applications for green invention patents has a large fluctuation relative to the number obtained, referring to the ideas of Hoang et al. and Pan et al., this paper uses the number of green invention patents obtained by enterprises plus 1 and takes the logarithm (LnGI) to measure the level of green technology innovation of enterprises [51,52].

4.2.4 Control variables.

Drawing on the ideas of Xu et al. and Ye et al., this paper selects other variables that may affect green development of enterprises from the provincial, industry and enterprise levels [9,11]. Firstly, the control variables at the provincial level include provincial economic development level (LnPgdp), provincial resource endowment (Rse) and provincial environmental regulation level (Evr). LnPgdp is measured by the logarithm of the real per capita GDP of the province. Rse is measured by (the sum of fixed asset investment in provincial agriculture and mining industry / fixed asset investment in the whole society of the province). Evr is measured by (the amount of investment completed in provincial industrial pollution control / provincial GDP × 100).

Secondly, the control variable at the industry level is the degree of industry competition (HHI). HHI is measured by the industry Herfindahl index, and the larger the value, the smaller the degree of competition in the industry.

Thirdly, the control variables at the enterprise level include energy consumption scale (LnTEC), enterprise size (LnSize), enterprise per capita assets level (LnCI), asset-liability ratio (Lev), return on equity (ROE), and ownership concentration (OC). LnTEC is measured by the logarithm of the total energy consumption of the enterprise. Consistent with the idea of obtaining corporate carbon emissions, the total energy consumption of enterprises is also calculated through the total energy consumption at the industry level. LnSize is measured by the logarithm of the number of employees in the enterprise. LnCI is measured by the logarithm of (corporate total assets / number of employees). Lev is measured by (total corporate liabilities / total corporate assets). ROE is measured by (corporate net profit / corporate net assets). OC is measured by the proportion of the largest shareholder’s shareholding. The data sources and expected directions of all variables in this paper are shown in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Names of variables and sources of data.

https://doi.org/10.1371/journal.pone.0316337.t002

4.3.1 Variable description.

In order to reduce the influence of extreme values, this paper carries out the Winsor tail reduction of the first 1% and the last 1% of all variables. The results of descriptive statistics of variables are shown in Table 3, where the difference between the maximum and minimum values of all variables and their average values is within a reasonable range. The minimum values of LnCEI and LnICT are not much different from their average and maximum values, indicating that the carbon emission intensity of enterprises in highly polluting industries is not too low and their ICT investment also has a certain scale.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Result of descriptive statistics of variables.

https://doi.org/10.1371/journal.pone.0316337.t003

4.3.2 Model test.

Firstly, the multicollinearity test of independent variable and control variables is carried out, and the VIF values of each variable are within 4, indicating that these variables do not have serious multicollinearity problems.

Secondly, the panel threshold effect test is carried out. In view of the fact that more than 50% of the sample size will be lost when the sample data is transformed into balanced panel data, the xtendothresdpd command is selected to estimate the dynamic panel threshold model of unbalanced panel data. At present, this command can only estimate the single threshold effect. Combined with the conclusion of the previous theoretical analysis, the single threshold model is reasonable. When conducting a single threshold existence test, only the lowest point of the blue curve can be found to indicate the existence of a single threshold effect. Fig 1 shows the threshold effect test results of the benchmark regression model, where the lowest point of the blue line curve is below the green horizontal line, indicating that the model has passed the single threshold effect test. The threshold value of LnICT is 12.3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Estimated value and confidence interval of threshold parameter in benchmark regression model.

Note: The blue line curve represents the likelihood ratio statistic (LR) of different threshold values γ, and the green horizontal line represents the critical value of LR at the 90% confidence level.

https://doi.org/10.1371/journal.pone.0316337.g001

5.2.1 Endogeneity test.

(1) Using the “Smart City” pilot policy for multi-period difference-in-difference (DID) model regression. The regression results of Column (1) in Table 5 show that the coefficient of PilotSC is significantly negative, indicating that the “Smart City” pilot policy can promote the green development of enterprises, which also supports the benchmark regression results of this paper. Fig 2 is the result of the parallel trend test. The results show that only in the second year after the actual policy occurred, the coefficient of PilotSC2 passed the 5% significance level test and was negative, indicating that the policy has passed the parallel trend test.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Parallel trend test of “Smart City” pilot policy.

https://doi.org/10.1371/journal.pone.0316337.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. The results of endogeneity test.

https://doi.org/10.1371/journal.pone.0316337.t005

1. (2) Using instrumental variable. In this paper, the topographic relief of the city where the enterprise is located is used as the instrumental variable of the enterprise’s ICT investment level, and then two-stage least squares regression (2SLS) is carried out [29]. From the relevant conditions, topographic relief is an important evaluation condition to determine whether a city is suitable for large-scale construction of ICT infrastructure, and the level of ICT infrastructure construction in a city will affect the level of ICT investment of enterprises. From the perspective of exogenous conditions, the carbon emission intensity of enterprises is mainly affected by the location advantage and green technology level of the city where they are located [53]. The location advantage of the city determines the development scale of the advanced manufacturing industry in the city, and the advanced manufacturing industry is often an industry with low carbon emission intensity. As a natural geographical variable, the topographic relief of the city is difficult to directly affect the carbon emission intensity of enterprises. Columns (2) in Table 5 show that instrumental variable passed the underidentification test (Kleibergen-Paap rk LM statistic) and the weak identification test (Cragg-Donald Wald F statistic). And the results show that the coefficient of LnICT is significantly negative, indicating that the benchmark regression results are robust.
2. (3) Adding variables that may be missed. Enterprises may be able to reduce carbon emission intensity after becoming pilot enterprises of carbon emission trading policy, so this paper sets the dummy variable PilotCE whether it is a pilot enterprise of carbon emission trading policy and adds it to the benchmark model to regress again. In addition, the carbon emission intensity of enterprises with environmental management system certification (ISO14001) may be lower, so this paper also sets the dummy variable PISO14 of whether the enterprise has ISO14001 certification to join the benchmark model and regress again. The results of columns (3) and (4) in Table 5 show that the coefficients of above_LnICT are significantly negative, which further verifies the benchmark regression results. Although the coefficient of below_LnICT in column (4) is also significantly negative, its absolute value is smaller than that of above_LnICT and its significance level is relatively lower, which is also basically in line with the conclusion of theoretical analysis.

5.2.2 Robustness test.

1. (1) Replace the dependent variable. The carbon emission intensity of enterprises in this paper only focuses on the carbon dioxide in the pollutant emissions of enterprises, and pays less attention to other pollutant emissions and waste treatment of enterprises. Therefore, this paper uses whether the enterprise is a key pollution monitoring enterprise (IKP) to re-measure the green development level of the enterprise. If the green development level of an enterprise is low, it is often a key pollution monitoring enterprise. On the contrary, it is often not a key pollution monitoring enterprise [54]. Because the dependent variable IKP is a binary dummy variable, the panel Logit model regression analysis is used. The regression results are shown in column (1) of Table 6, where the coefficient of LnICT is significantly negative.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. The results of robustness test.

https://doi.org/10.1371/journal.pone.0316337.t006

1. (2) Replace the independent variable. This paper uses two methods to re-measure the level of corporate ICT investment. The first method is to use the degree of digital transformation of enterprises, because the degree of digital transformation can also reflect the overall ICT investment level of the enterprise to a certain extent. The degree of digital transformation of enterprises (LnWordD) is measured by adding 1 to the total number of word frequencies related to digital transformation in the annual report of the enterprise and then taking the logarithm. Considering that the word frequency data is count data, the threshold effect is difficult to measure, and this paper uses the two-way fixed effect model to test. The second method is to use the relative level of corporate ICT investment. The absolute level of corporate ICT investment used above may not reflect the scale of corporate ICT investment relative to other types of investment. The relative level of corporate ICT investment (ICTD) is equal to the ratio of total corporate ICT investment to total corporate assets. The regression results are shown in columns (2) and (3) of Table 6, where the coefficients of LnWordD and above_ICTD are significantly negative, indicating that the baseline regression results are robust.

5.3.1 Mechanism of production efficiency.

Columns (1) and (2) in Table 7 show the regression results with LnTFP as the dependent variable, where Column (1) does not control the year fixed effect. The results show that the coefficients of LnICT are significantly positive, indicating that ICT investment can improve the production efficiency of enterprises in high-pollution industries. At the same time, improving production efficiency can reduce resource consumption intensity, control excessive emissions of pollutants, and promote green development of enterprises [39]. In summary, production efficiency is indeed one of the influencing mechanisms, which verifies Hypothesis 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. The results of mechanism test.

https://doi.org/10.1371/journal.pone.0316337.t007

5.3.2 Mechanism of green technology innovation.

Column (3) and (4) in Table 7 shows the regression results of LnGI as a dependent variable, where Column (3) does not control the year fixed effect. The results show that the coefficients of LnICT are significantly positive, indicating that ICT investment can improve the level of green technology innovation of enterprises in high-pollution industries. At the same time, green technology innovation can promote green development of enterprises by improving production processes and technologies and designing more environmentally friendly products [41]. In summary, green technology innovation is indeed one of the influencing mechanisms, which verifies Hypothesis 3.

5.4.1 Different management levels.

Referring to the idea of Qiu & Yu, the management level is obtained by regression of enterprise management cost to enterprise scale, enterprise export volume and enterprise price markup [55]. The absolute value of the residual of the regression is used to measure the management level of the enterprise, and the smaller the absolute value of the residual, the higher the management level. According to the median value of the sample management level, the high pollution industry enterprises are divided into enterprises with lower management level and enterprises with higher management level. The management level will affect the application efficiency of ICT products. Enterprises with low management level are difficult to integrate ICT products into development strategies and achieve green development [56]. However, enterprises with higher management level can organize employees to learn cutting-edge technology products and unified service standards in time, and promote cutting-edge technology products to play a significant role [57].

Columns (1) and (2) in Table 8 show the regression results of enterprises with different management levels. In the enterprise group with lower management level, the coefficients of below_LnICT and above_LnICT are not significant. In the enterprise group with higher management level, the coefficient of above_LnICT is significantly negative. To sum up, ICT investment has a significant impact on green development of enterprises with higher management level and has no significant impact on the green development of enterprises with lower management level.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. The results of heterogeneity test.

https://doi.org/10.1371/journal.pone.0316337.t008

5.4.2 Different industry characteristics.

According to the different production and technical characteristics of the industry, enterprises in high-pollution industries can be divided into light industrial enterprises and heavy industrial enterprises. Industry characteristics will affect the purpose of enterprise ICT investment. Light industrial enterprises are mostly labor-intensive enterprises, and they are more committed to using ICT products to integrate supply chains, which has a relatively small impact on energy conservation and emission reduction [58]. And they mainly reduce carbon emissions by introducing new production equipment or pollutant treatment equipment, which eventually leads to the insignificant impact of ICT investment on green development of enterprises. Heavy industrial enterprises are mostly resource-intensive enterprises and have a high demand for advanced technology. Thus, they emphasize the use of ICT products to achieve scale manufacturing and agile manufacturing [11], which will profoundly affect the energy conservation and emission reduction behavior of enterprises.

Columns (3) and (4) in Table 8 show the regression results of enterprises in different industries. In the light industrial enterprise group, the coefficients of below_LnICT and above_LnICT are not significant. In the heavy industry enterprise group, the coefficient of above_LnICT is significantly negative. To summarize, ICT investment has a significant impact on green development of heavy industrial enterprises, and has no significant impact on the green development of light industrial enterprises.

5.4.3 Different ownership nature.

According to the nature of ownership, enterprises in high-polluting industries are divided into non-state-owned enterprises and state-owned enterprises. The difference in the nature of ownership will also affect the purpose of ICT investment. ICT products can better achieve enterprise carbon emission reduction under the premise of low cost. Non-state-owned enterprises have a high degree of marketization, and they often allocate resources according to the principle of profit maximization [59]. Therefore, the purpose of ICT investment of non-state-owned enterprises includes promoting green development of enterprises. The degree of marketization of state-owned enterprises is relatively low and their resources are not necessarily allocated according to the principle of profit maximization [60]. They do not necessarily take the most cost-effective way, but to achieve green transformation by introducing new production equipment and pollutant treatment equipment. And their main purpose of ICT investment may be to improve economic benefits. This leads to the fact that there is not necessarily a causal relationship between ICT investment and carbon emission intensity of state-owned enterprises.

Columns (5) and (6) in Table 8 show the regression results of enterprises of different ownership types. In the non-state-owned enterprise group, the coefficients of below_LnICT and above_LnICT are significantly negative. In the state-owned enterprise group, the coefficients of below_LnICT and above_LnICT are not significant. In summary, ICT investment has a significant impact on green development of non-state-owned enterprises, and has no significant impact on the green development of state-owned enterprises. The reason why the coefficient of below_LnICT in column (5) is also significant may be that ICT investment can increase the degree of asset-lightening of non-state-owned enterprises compared with state-owned enterprises, and the asset-light model is conducive to green technology innovation and green development.

5.4.4 Different ICT investment types.

2016 is the year in which digital technology began to play a major role in China’s economy and society compared with traditional ICT. Compared with traditional ICT, digital technology has stronger substitutability and synergy to traditional production factors, and its impact on green development of enterprises is greater in scope and intensity [61]. Therefore, this paper divides the samples into two groups according to the year before 2016 and after 2016 to test the impact of traditional ICT investment and digital technology investment.

Columns (7) and (8) in Table 8 show the regression results of different ICT investment types. Since the span of each time period after segmentation is short, neither of these two regressions controls the year fixed effect. The results show that, in the traditional ICT investment group, the coefficient of below_LnICT is not significant and the coefficient of above_LnICT is significantly negative. In the digital technology investment group, the coefficients of below_LnICT and above_LnICT are significantly negative. In conclusion, digital technology investment has a greater impact on green development of enterprises in high-pollution industries than traditional ICT investment, because it does not have to cross the threshold to promote green development of enterprises. In Column (8), the absolute value of the coefficient of below_lnICT is larger than that of above_lnICT. The possible explanation is that the network effect of digital technology investment is greater and the energy consumption of ICT products is still high at this time. When the level of ICT investment is low, the carbon emission intensity of enterprises will be relatively low.