Smart City, Digitalization and CO₂ Emissions: Evidence from 353 Cities in China

Abstract

The development of digital technology provides new governance methods for achieving the goal of “carbon peaking and carbon neutrality”. Since 2013, the pilot construction of smart cities in China has strengthened the government’s digital governance capabilities and significantly influenced the reduction in carbon emissions. This paper provides empirical evidence for the driving effect of digitization on carbon emission reduction based on panel data from 353 cities in China. The results show that digital governance based on smart city construction pilots has significantly reduced regional carbon emissions, and the implementation of smart city construction pilots has reduced regional carbon emissions by an average of 6.6%, and this effect is sustainable over the long term. The increase in the level of digitalization has significantly promoted carbon emission reduction. From the perspective of the impact path, regional green patent innovation has played a significant partial intermediary effect in the process of digitization-driven carbon emission reduction. From a micro-mechanism standpoint, digitization plays a significant role in promoting the green innovation of high-polluting listed companies.

1. Introduction

The Paris Agreement, approved at the Paris Climate Conference in 2015, proposes to control the global average temperature rise within 2 °C by 2100, and strives to control it within 1.5 °C. At the 75th United Nations General Assembly, as one of the roughly 200 signatory states, China committed to the international community about carbon neutrality by 2030 and carbon peaking by 2060, which was included in the 2021 Chinese government work report. According to the assessment from the Intergovernmental Panel on Climate Change (IPCC), the cumulative global CO₂ emissions from 2011 to 2100 must be limited to 400 billion tons to control the temperature rise of 1.5 °C with a likelihood greater than 66% [1]. Based on a detailed assessment, China’s carbon emissions must be cut to between 3 billion and 8 billion tons in 2030 and between 1 billion and 2 billion tons in 2050 to meet the target [2]. Accelerating the realization of the goal of “carbon neutrality and carbon peaking “ has become a top priority for the Chinese government, attracting the attention of all sectors of society. Due to the high cost and low return of carbon emission reduction, companies and individuals that emit carbon lack the will to reduce emissions actively. Carbon emission reduction has the problem of “market failure”; consequently, achieving the goals of carbon emission reduction requires the comprehensive application of various governance means and the development of a modern governance capacity and governance system.

With the development and popularization of digital technology, the smart city as a new model of city governance has drawn global interest, and countries around the world have initiated a construction boom of smart cities. The Singapore government implemented the “Intelligent Nation 2015” plan in 2006; the South Korean government launched the smart city construction plan called “U-city” in 2006; the Netherlands started “Smart City and Planning and Construction for Amsterdam” in 2011; IBM officially proposed “Smarter Cities” in 2010, and built the world’s first unified urban data management center in Rio de Janeiro, Brazil. China’s smart city construction plan was initiated in 2012, and 277 locations (including cities, counties, and towns) are currently designated as pilot smart city construction sites. The pilot construction of smart cities led by local governments has significantly promoted digital governance capabilities, smart application projects, such as lighting, smart transportation, smart environmental protection, smart energy, and smart factories are gaining popularity, all of which have had a significant impact on environmental pollution reduction in cities. It is essential to scientifically evaluate the role of digital governance promoted by smart city construction in the process of carbon emission reduction, and maximize the benefits of digital technology to accelerate carbon emission reduction for timely achievement of “carbon neutrality and carbon peaking” on time.

Aiming at the literature gaps, this study explores the association between digitalization and CO₂ emissions based on the policy of the “smart city” construction pilots in China. The marginal contributions of the present study are in the following three aspects. Firstly, with the help of the quasi-natural experiments about the policy of smart city construction pilots, this paper provides convincing evidence for the positive role that digital governance plays in restraining regional carbon emissions. Secondly, to alleviate the endogenous problem, we take “smart city pilots” and “broadband China pilots” as instrumental variables of the level of digitization to estimate the driving effect of digitization on carbon emission reduction. Thirdly, based on the matching of the green innovation micro-data of high-polluting listed companies with the data of the digitization level of their regions, this paper explains the micro-governance mechanism for digitization-driven carbon emission reduction.

The remainder of this study is organized as follows. Section 2 presents the relevant literature about the digital economy and CO₂ emissions. Section 3 is materials and methods. Section 4 is the empirical results and analysis, a quasi-natural experiment is used to test the impact of the implementation of smart city pilot policies on regional carbon emissions, and this paper also analyzes the impact of the digitization level on CO₂ emissions. Section 5 further explores the mechanisms by which digitalization drives regional CO₂ emissions reductions. Section 6 provides conclusions and potential policy implications.

2. Literature Review

2.1. Influencing Factors of CO₂ Emissions

An important prerequisite for achieving the “carbon peaking and carbon neutrality” goal is to clarify the causes of carbon emissions. After Grossman and Krueger proposed and empirically tested the environmental Kuznets curve hypothesis [3], many scholars began to explore the impact of economic development on carbon emissions, and found an inverted U-shaped relationship between CO₂ emissions and per capita income, and called it the Carbon Kuznets Curve [4]. The research conclusions of Ang; Apergis, and Payne; and Jalil and Mahmud based on the data from France, six Central American countries, and China, respectively, further supported the Carbon Kuznets Curve hypothesis [5,6,7]. Secondly, urbanization is considered to be another important cause of carbon emission problems, but the research conclusions on this topic are inconsistent. Satterthwaite studied urban environmental problems in developing countries and found that environmental pollution tends to ease with the expansion of the urban scale [8]. Martinez-Zarzoso and Maruotti found an inverted U-shaped relationship between urbanization and carbon emissions [9]. In addition, some scholars have paid attention to other factors affecting carbon emissions, such as Chen et al. who studied the impact of trade opening on carbon emissions [10]. Jalil et al. studied the impact of financial development on carbon emissions [11].

2.2. Governance, Market Mechanism and CO₂ Emissions

How to build a modern governance capacity and governance system for reducing carbon emissions is an important issue that all countries and governments are highly concerned about. It is also a research hotspot in academia, mainly involving research on governance by government regulation and market mechanisms.

The government’s environmental regulations are commonly used by governments around the world to reduce carbon emissions. Danish et al. suggested that governance reduced carbon emissions in BRICS countries from 1996 to 2017 [12]. Güney believed that the improvement in governance promotes the investment and long-term planning for solar energy, which reduces carbon emissions [13]. Khan et al. believed that effective governance is one of the critical factors in reducing carbon emissions in Morocco [14]. However, Zhang et al. believed that the development of China is still in the “green paradox” stage, so there is an inverted “U” curve between environmental regulation and carbon dioxide emissions [15].

The role of the market trading mechanism in carbon emission reduction is increasingly significant, which has been supported by more and more of the literature. Metcalf believed that, from the perspective of the effect of pollution control in the United States, the carbon price policy based on the market mechanism is better than the “command and control” policy [16]. Borenstein et al. argued that the design of environmental market mechanisms should take into account the ex-ante uncertainty of carbon emission reductions [17]. Regarding the implementation effect of China’s market trading policy on pollution emission, the conclusions of the research literature are different, such as Liao et al. and Cheng et al. who demonstrated that the emission trading policy has effectively reduced carbon emissions in Guangdong and Shanghai [18,19]. Qian et al. argued that a reasonable design of the carbon emission rights distribution mechanism among regions is conducive to the realization of China’s greenhouse gas peaking goal and the optimization of emission reduction costs [20].

2.3. Digital Economy and Carbon Emissions

The development of the digital economy has provided new ideas for solving environmental problems, and has become a hot spot of academic attention in recent years. Li et al. argued that the development of the digital economy in Chinese cities significantly reduces PM_2.5 [21]. Zhou et al. suggested that the development of the digital economy has significantly reduced smog pollution in China [22]. Shi et al. believed that China’s smart city pilots have significantly reduced pollution emissions [23]. Specifically, in terms of research on the impact of the digital economy on carbon emissions, the research conclusion of most scholars supported the hypothesis that the digital economy is beneficial to curbing carbon emissions. Representative studies include that of Zhang et al. who argued that the development of the digital economy improves the performance of carbon emission reduction by affecting energy intensity, the energy consumption scale, and urban afforestation [24]. Ma et al. argued that digitization helped China achieve low-carbon growth [25]. Ren et al. suggested that internet development has accelerated the decline in energy consumption intensity through economic growth, R&D investment, human capital, financial development, and the upgrading of industrial structure [26]. However, some studies have inconsistent conclusions, such as Li et al.; Wang; and Li and Wang, who believed that there is an inverted U-shaped relationship between the digital economy and carbon emissions: the development of the digital economy initially increases carbon emissions, and then reduces carbon emissions with the further development of the digital economy [27,28,29].

In addition, the impact of digital technology and digital finance on carbon emissions has been the focus of the literature in recent years. Liu et al. believed that the development of digital technology can not only reduce local carbon emissions but also promote carbon emissions in surrounding areas [30]. Wang et al. suggested that technological innovation in the information industry will increase carbon emission intensity, while cross-industry technology spillovers can sustainably reduce carbon emission intensity in the long run [31]. The conclusion of Zhao’s research based on Chinese provincial panel data supported the hypothesis that digital finance has a significant inhibitory effect on carbon emissions [32]. Liu et al. argued that digital finance can alleviate financial constraints and increase R&D investment to promote green innovation [33]. Zhang et al. argued that the synergistic effect of digital finance and green technology innovation can significantly improve the efficiency of local carbon emissions [34].

2.4. Literature Gap

Although the existing related research literature has made efforts to focus on the causes of carbon emissions and the governance of carbon emissions, some literature gaps still exist. Firstly, existing empirical studies have not paid enough attention to the endogenous issues between digitization and carbon emission reduction, and the driving effect of digitization on carbon emission reduction needs to be supported by more credible empirical evidence. Secondly, the existing research does not explain the micro-mechanism of digital-driven carbon emission reduction in depth and lacks data support at the micro-enterprise level.

3. Materials and Methods

3.1. Hypotheses

A large number of published documents have studied the influencing factors of carbon emissions, and some scholars have also paid attention to the impact of the digital economy on carbon emissions in recent years. However, the relationship between the digital economy and carbon emissions is still unclear, and the impact mechanism needs to be further clarified. China’s smart city pilot construction provides a good quasi-natural experiment for studying the inhibitory effect of digital governance on carbon emissions. Since 2012, the plan for China’s smart city construction has been launched in hundreds of cities; technology has been vigorously applied and promoted, significantly improving the digital level of cities and providing digital technology support for carbon emission governance. Based on existing research literature, this paper discusses the impact of digitalization on carbon emissions and its mechanism from the perspective of smart cities, and puts forward the following hypotheses:

Hypothesis 1. 

The implementation of smart city policy and the digitalization it promotes have a significant negative impact on urban carbon emissions, namely, digital governance significantly curbs carbon emissions in Chinese cities.

Hypothesis 2. 

Green innovation has played an important intermediary role, digitalization can reduce carbon emissions by improving green innovation capability. From a micro perspective, digitalization affects carbon emissions by influencing the patent innovation capability of enterprises.

3.2. Econometric Model and Method

Firstly, the research takes the shock of the smart city pilot policy as a quasi-natural experiment of digitalization, constructs an experimental group (smart city construction pilot city) and a control group (non-smart city construction pilot city), and adopts the difference-in-difference method (DID) to estimate. Referring to the literature by Beck et al. [35], the parallel trend test is firstly carried out; that is: if there is no digital-driven impact of smart city construction pilots, there should be no systematic differences over time between the carbon emissions of smart city construction pilots and those of non-pilot cities. The model is set as follows:

$$\begin{array}{c}LnC{O}_2\_cit{y}_{it}=c+{\displaystyle \sum _{\mathrm{m}=-11}^{-1}{\alpha }_m\times Digital\_di{d}_{i,t+m}}+{\alpha }_0\times Digital\_di{d}_{i,t}\\ +{\displaystyle \sum _{m=1}^4{\alpha }_m\times Digital\_di{d}_{i,t+m}}+\beta \times X_{it}+{\theta }_i+{\gamma }_t+{\epsilon }_{it}\end{array}$$

(1)

where LnCO_{2_}city_it denotes the logarithm of the carbon emissions of the city i in year t, the 4 leads and 11 lags of Digital_did refer to the implementation of the smart city pilot policy at time t + m. In addition to the parallel trend test, according to the study by Beck et al. and Moretti [35,36], this model can be used to test the dynamic effects of the smart city pilot policy, the coefficients on the lead terms, α₁ through α₄, allow us to determine how the carbon emissions of a city in a given year respond to future changes in the policy. The coefficients on the lag terms, α₋₁₁ through α₋₁, allow us to examine how the implementation of the smart city pilot policy propagates over time, and in particular, whether the effect is permanent.

The multi-period DID benchmark regression model is set as follows:

$$LnC{O}_2\_cit{y}_{it}=c+{\alpha }_1\times Digital\_di{d}_{it}+{\beta }_1\times X_{it}+{\theta }_i+{\gamma }_t+{\epsilon }_{it}$$

(2)

where Digital_did_it represents whether the city i starts the pilot construction of a smart city in year t, X_it is the vector combination of control variables affecting carbon emissions, γ_t is the year fixed effect, θ_i is the city fixed effect, ε_it is the random error term.

Furthermore, this paper estimates the impact of digitalization on carbon emissions with the panel data of cities in China, and the estimation model is set as follows:

$$LnC{O}_2\_cit{y}_{it}=c+{\alpha }_2\times Digital\_inde{x}_{it}+{\beta }_2\times X_{it}+{\theta }_i+{\gamma }_t+{\epsilon }_{it}$$

(3)

where Digital_index_it denotes the digitization level, X_it denotes the other control variables that affect carbon emissions, γ_t is the year fixed effect, θ_i is the city fixed effect, and ε_it is the random error term.

Finally, to investigate the impact mechanism, this paper firstly tests the mediating role of green patent innovation in the process of digitalization affecting carbon emission reduction from the macro-city sample level. The mediation effect test process is carried out in three steps. The estimation model of the first step uses model (3), and the test models of the second and third steps are set as follows:

$$Green\_paten{t}_{it}=\mathrm{c}+{\alpha }_3\times Digital\_inde{x}_{it}+{\beta }_3\times X_{it}+{\epsilon }_{it}$$

(4)

$$LnC{O}_2\_cit{y}_{it}=\mathrm{c}+{\alpha }_4\times Green\_paten{t}_{it}+{\alpha }_5\times Digital\_inde{x}_{it}+{\beta }_4\times X_{it}+{\epsilon }_{it}$$

(5)

where Green_patent_it is the mediating variable.

According to the public data of the Chinese government, China’s top three carbon-emitting industries (coal-fired power plants, steel, and cement) accounted for more than 60% of the country’s total emissions in 2020, the total carbon emissions of the top 100 listed companies were 4.424 billion tons, accounting for about 44.7% of the national total. These data show that reducing carbon emissions of high-pollution industries is the focus of urban carbon emission reduction. Therefore, this paper discusses the significant impact of digitalization on green patent innovation of listed enterprises in high-pollution industries, providing micro evidence for the intermediary role of green innovation. Referring to the research method of Acemoglu and Restrepo [37], based on the matching of the green patent data of high-polluting listed companies with the digitalization level of the region where they are located, this paper explores the micro-mechanism of digitalization driving green innovation of high-polluting listed companies. The model is as follows:

$$\begin{array}{ll}Firm\_greenpaten{t}_f& =c+{\alpha }_6\times Pollute\_dumm{y}_f+{\alpha }_7\times Digital\_inde{x}_i+\\ & {\alpha }_8\times Pollute\_dumm{y}_f\times Digital\_inde{x}_i+{\beta }_5\times Y_f+{\epsilon }_f\end{array}$$

(6)

where Firm_greenpatent_f is the number of green patent applications of listed company f, Pollute_dummy_f is a dummy variable of whether listed company f belongs to a highly polluting industry, which is constructed based on whether the listed company belongs to the 16 high-pollution industries. Y_f denotes the other factors affecting the green patent innovation of listed companies.

3.3. Variables and Data Resources

The explained variable is CO₂ emissions. The CO₂ emission data for each city comes from CEADs, which uses two sets of nighttime light data (DMSP/OLS and NPP/VIIRS data) provided by the National Geophysical Data Center to invert the CO₂ emissions of 2735 counties in China from 1997 to 2017 [38], the variable is named: CO₂_cityI. The carbon emission data from CEADS is the authoritative data widely used in relevant research, so it is used as the benchmark regression data. However, the shortcoming of these data is that they are only updated to 2017. To make up for this deficiency, this paper also uses the method of Wu and Guo (2016) to obtain data that can be updated to 2020 [39], which is used as a substitute variable for robustness testing. According to the CO₂ emission coefficient provided by IPCC2006, the sum of CO₂ emissions generated by electricity, coal gas, liquefied petroleum gas, and thermal energy consumption in each region is calculated, respectively, which is used as an alternative indicator, named: CO₂_cityII. The original data come from the “China Urban Statistical Yearbook” and the “China Urban Construction Statistical Yearbook” (2006–2020).

The pilot construction of smart cities, designed at the top level of the state and led by local city governments, can be used as a quasi-natural experiment for local governments’ digital governance capabilities. The State Council of China successively implemented 90, 103, and 84 smart pilots (including cities, counties, and towns) in January 2013, August 2013, and April 2015, respectively, involving 155 cities, including eastern cities, as well as central and western cities and northeastern cities. This study takes 155 pilot smart cities as the processing group for digital governance and other non-pilot cities as the control group, to construct a Differences-in-Differences estimation variable of digital governance, named: Digital_did. According to the availability of relevant data about digitization in China, the number of internet broadband accesses per capita, the percentage of digital employees in the population, and the per capita telecommunication business income are used as sub-indicators to construct a comprehensive regional digitization index. The first comprehensive digitization index is obtained using the arithmetic mean method, named: Digital_indexI, the specific methods are as follows: Step 1 is to standardize the sub-indicator data; Step 2 is to calculate the mean value of the standardized value, that is, subjectively, all sub-indicators are given the same weight. The second comprehensive digitization index is obtained using the principal component analysis method (Hotelling, 1933), named: Digital_indexII, the specific methods are as follows: Step 1 is to standardize the sub-indicator data; Step 2 is to perform principal component analysis on the standardized indicators, determine the number of principal components according to the cumulative variance contribution rate, and calculate the score of each principal component. Step 3 is to calculate the weight according to the proportion of the variance contribution rate of each principal component in the cumulative variance contribution rate of the extracted principal component, and then obtain the comprehensive index by weighted average. The original data come from the “China Urban Statistical Yearbook”.

The mediating variable is green innovation. The city’s green innovation data adopt the number of green patent applications in each city, including the sum of the number of green invention applications and the number of green utility model applications, and the variable is named: Green_patent.

The explained variable in the micromechanism test is the green innovation of listed companies. The green innovation of listed companies is quantified by the number of green patent applications (the sum of the number of green inventions and green utility models submitted independently and jointly), and the variable is named: Firm_greenpatent. The original patent data come from China’s national intellectual property patent database, and green patents are identified according to the “Green Patent List” issued by the World Intellectual Property Organization (WIPO) in 2010.

According to the research literature, the influencing factors of carbon emissions mainly include the level of economic development [6], industrial structure [40], and openness [10]. Therefore, the empirical test controls the impact of per capita GDP (Gdp_per_capita), the ratio of secondary and tertiary industry output to GDP (Industry), and the ratio of foreign investment to GDP (Fdi) on regional CO₂ emissions. The original data come from the “China Statistical Yearbook” and “China Urban Statistical Yearbook”. At the same time, in the estimation based on micro-listed company data, company characteristics such as years of establishment (Open_time), net profit of total assets (Profit), management shareholding ratio (Manage), and equity nature (Private_dummy) are also controlled. The original data come from the Wind database of China.

Figure 1 shows the changing trend of total carbon emissions in 353 cities in China from 2001 to 2017. It can be seen that the growth trend of carbon emissions in Chinese cities continued from 2001 to 2014; however, since 2014, this growth trend has ceased. The statistical description of explained variables, core explanatory variables, and mediating variables is shown in

Table 1. Statistical description of variables.

Variables	Obs	Year	Mean	Std. Dev.	Min	Max
LnCO2_cityI	6001	2001–2017	2.495	1.131	−2.175	5.441
Digital_did	6001	2001–2017	0.090	0.286	0.000	1
Digital_indexI	4950	2003–2020	0.112	2.681	−3.765	32.495
Digital_indexII	4950	2003–2020	0.043	1.022	−1.462	12.083
Green_patent	4125	2003–2017	139.0	497.0	0.000	9080
Firm_greenpatent	30,602	2003–2020	2.106	18.486	0.000	1532

Note: LnCO₂_cityI is the logarithm of CO₂_cityI.

.

4. Empirical Results and Analysis

4.1. Quasi-Natural Experiment Analysis: Smart City Pilot and Carbon Emission Reduction

4.1.1. Parallel Trend Test and Dynamic Effect Estimation

China started implementing the policy of smart city pilot in 2013 and 2015, so 2013 is set as the 0 period of policy implementation. At the same time, according to the availability of data, the time window for the test is 2001–2017, including 11 periods before the policy was implemented (one period was removed to solve the collinearity problem) and four periods after the policy was implemented. According to the empirical model (1), the estimated results of dynamic effects are shown in column (1) of

Table 2. Estimated results of dynamic effects of smart city pilots on carbon emissions.

	(1)	(2)	(3)
VARIABLES	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI
Digital_did_pre_11	−0.005	−0.005	0.010
	(0.007)	(0.008)	0.008
Digital_did_pre_10	0.002	0.003	0.021 *
	(0.010)	(0.010)	0.011
Digital_did_pre_9	0.008	0.010	0.026 *
	(0.012)	(0.013)	0.013
Digital_did_pre_8	0.011	0.013	0.028
	(0.017)	(0.018)	0.017
Digital_did_pre_7	0.016	0.019	0.034
	(0.019)	(0.021)	0.020
Digital_did_pre_6	0.012	0.014	0.018
	(0.022)	(0.024)	0.021
Digital_did_pre_5	0.008	0.009	0.015
	(0.024)	(0.026)	0.023
Digital_did_pre_4	0.007	0.010	0.017
	(0.024)	(0.026)	0.023
Digital_did_pre_3	−0.003	−0.001	0.014
	(0.025)	(0.027)	0.025
Digital_did_pre_2	−0.042	−0.045	−0.023
	(0.029)	(0.032)	0.028
Digital_did_pre_1	−0.049	−0.052	−0.033
	(0.030)	(0.032)	0.029
Digital_did_current	−0.090 ***	−0.094 ***	−0.074 **
	(0.032)	(0.034)	0.031
Digital_did_post_1	−0.089 ***	−0.092 ***	−0.073 **
	(0.032)	(0.034)	0.031
Digital_did_post_2	−0.082 **	−0.085 **	−0.068 **
	(0.032)	(0.034)	0.032
Digital_did_post_3	−0.080 **	−0.082 **	−0.069 **
	(0.032)	(0.035)	0.032
Digital_did_post_4	−0.104 ***	−0.107 ***	−0.087 **
	(0.035)	(0.037)	0.034
Gdp_per_capita			0.004
			0.005
Fdi			−0.003 ***
			0.000
Industry			0.004 ***
			0.001
Constant	1.725 ***	1.728 ***	1.461
	(0.011)	(0.012)	0.057
Observations	6001	5304	6001
R-squared	0.898	0.894	0.901

Note: ***, **, and * denote significance at the levels of 1%, 5%, and 10%, respectively. The numbers in parentheses are standard errors.

, and Figure 2 is the trend diagram of the dynamic effect. The estimated coefficients on the lag terms (α₋₁₁ through α₋₁) are not significant, which satisfies the parallel trend assumption. The estimated effects of the current and lead terms (α₀ through α₄) are all negative, at the level of 1%, 1%, 5%, 5%, and 1%, respectively, indicating that the dynamic effect of the policy is significant, and the effect of the smart city pilot on the reduction in carbon emissions will be sustained in the next terms. The estimates in column (2) of Table 2 exclude the city samples that started the pilot program in 2015, and the results are basically the same as those in column (1). Column (3) of Table 2 is the estimated result after adding control variables, which is consistent with the results of columns (1) and (2), and the dynamic effect is also significant.

4.1.2. Baseline Regression: Multi-Period DID Estimation

The baseline regression uses the panel data of 353 cities from 2001 to 2017 and the empirical model (2), and the explained variable is the logarithm of carbon emission I (LnCO₂_cityI).

Table 3. Impact of smart city pilot on carbon emissions (multi-period DID).

	(1)	(2)	(3)	(4)
VARIABLES	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI
Digital_did	−0.068 ***	−0.070 ***	−0.067 ***	−0.066 ***
	(0.018)	(0.018)	(0.018)	(0.017)
Gdp_per_capita		0.001	0.001	0.002
		(0.005)	(0.005)	(0.005)
Fdi			−0.003 ***	−0.003 ***
			(0.000)	(0.000)
Industry				0.004 ***
				(0.001)
Constant	2.930 ***	2.926 ***	2.929 ***	2.607 ***
	(0.015)	(0.021)	(0.021)	(0.073)
Year_fix	yes	yes	yes	yes
City_fix	yes	yes	yes	yes
Observations	6001	6001	6001	6001
R-squared	0.896	0.896	0.898	0.900

Note: *** denotes significance at the levels of 1%. The numbers in parentheses are standard errors.

presents the results of the multi-period DID estimation. From columns (1) to (4), the Gdp_per_capita, Fdi, and Industry are sequentially added as control variables. The estimates range from −0.068 to −0.066, at the level of 1%. The estimation results show that the policy of smart city pilot construction has significantly reduced regional carbon emissions.

4.1.3. Robustness Test

This research carried out a robustness test from the following three aspects: Firstly, the propensity score matching method (PSM) was used to match the treatment and control groups with the closest characteristics before DID estimation to reduce the effect of differences between groups. Secondly, this paper re-estimated by replacing the explained variable indicators and time windows to alleviate the impact of indicator measurement errors and time window selection. Finally, referring to the placebo test method of Chetty et al. [41], the smart city pilot cities and implementation years were randomly selected for estimation to test the impact of omitted variables.

Figure 3 and Figure 4 report the results of the common support assumption test. It can be seen intuitively from Figure 3 that the standardization deviation of all variables was reduced after matching. It can be seen intuitively from Figure 4 that only a small number of samples were lost in the distribution of propensity scores. Most observations are on support, and only a few are off support (untreated: 56; treated: 55). Therefore, PSM-DID was introduced for the robustness test.

Table 4. Robustness test results using the PSM-DID model.

	(1)	(2)	(3)	(4)
VARIABLES	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI
Digital_did	−0.066 ***	−0.068 ***	−0.068 ***	−0.066 ***
	(0.018)	(0.018)	(0.018)	(0.017)
Gdp_per_capita		0.002	0.002	0.002
		(0.003)	(0.003)	(0.005)
Fdi			0.003	−0.003 ***
			(0.004)	(0.000)
Industry				0.004 ***
				(0.001)
Constant	2.945 ***	2.937 ***	2.935 ***	2.607 ***
	(0.015)	(0.018)	(0.018)	(0.073)
Year_fix	yes	yes	yes	yes
City_fix	yes	yes	yes	yes
Observations	5890	5890	5890	6001
R-squared	0.901	0.901	0.901	0.900

Note: *** denotes significance at the levels of 1%. The numbers in parentheses are standard errors.

presents the estimation results of PSM-DID, and the estimates range between −0.066 and −0.068 depending on the set of controls, at the level of 1%, which are consistent with the DID estimation results in Table 3. The above results show that the estimation results after changing the model are very robust, further supporting the hypothesis that “smart city pilots significantly promote carbon emission reductions”.

The empirical estimation uses CO₂_cityII as an alternative indicator for CO₂_cityI; according to the availability of data, the time window of panel data is 2006–2020. The empirical estimation results are shown in

Table 5. Robustness test results for changing the explained variable and time window.

	(1)	(2)	(3)
VARIABLES	LnCO2_cityII	LnCO2_cityII	LnCO2_cityII
Digital_did	−0.063 ***	−0.063 ***	−0.061 ***
	(0.020)	(0.020)	(0.020)
Gdp_per_capita	0.010 **	0.010 **	0.009 **
	(0.002)	(0.003)	(0.003)
Fdi		−0.000	−0.000
		(0.001)	(0.001)
Industry			0.006 ***
			(0.001)
Constant	6.353 ***	6.351 ***	5.739 ***
	(0.026)	(0.029)	(0.110)
Year_fix	yes	yes	yes
City_fix	yes	yes	yes
Observations	4095	4095	4095
R-squared	0.353	0.353	0.358

Note: *** and ** denote significance at the levels of 1% and 5%, respectively. The numbers in parentheses are standard errors.

. The estimates in columns (1) to columns (3) range between −0.061 and −0.063 at the level of 1% depending on the set of controls, which are consistent with the estimation results in Table 3. The values of estimates changed due to changes in the construction of the explained variable indicators, which were in line with expectations. Therefore, the estimation results are still robust after changing the explanatory variable data and time window.

Although the fixed effects of city and year have been controlled, other unobservable regional feature variables may have different effects over time, and a direct placebo test was used to alleviate the estimation bias caused by omitted variables. This study randomly selected smart city pilots and implementation years to repeat 1000 regressions. The kernel density of the estimated T value and the estimated coefficients are shown in Figure 5 and Figure 6, respectively. The results show that the T value and estimates based on the random sample test are all normally distributed around 0. The estimated coefficient of the DID baseline regression in Table 3 is −0.066, which is basically independent of the distribution shown in Figure 6. The results of the placebo test show that the omitted variables have little effect on the estimated results, and the estimated results are very robust.

Based on the above estimates, the estimated coefficients of the core explanatory variable in column (4) of Table 3 and Table 4 are the same (−0.066), the estimated coefficient in column (4) of Table 5 is −0.061, which can be explained by the possibility that the constructed carbon emission substitute index does not fully cover all carbon emissions due to the availability of data and the limitations of methods; the estimated coefficient is in line with expectations. Therefore, according to the estimation model, the estimation results can be interpreted as: the implementation of the smart city construction reduced carbon emissions by about 6.6% on average.

4.2. Estimation Results and Analysis Based on the Panel Fixed Effects: The Impact of Digitalization on Carbon Emissions

4.2.1. Baseline Regression

The baseline regression uses panel data of cities from 2003 to 2017 according to the availability of digital index data, and the empirical model is model (3).

Table 6. Baseline regression results of the impact of digitization on carbon emissions.

	(1)	(2)	(3)	(4)
VARIABLES	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI
Digital_indexI	−0.009 ***	−0.015 ***	−0.015 **	−0.013 **
	(0.003)	(0.002)	(0.007)	(0.007)
Gdp_per_capita		0.005 ***	0.005	0.006
		(0.001)	(0.006)	(0.005)
Fdi			−0.285	−0.303
			(0.189)	(0.193)
Industry				0.006 ***
				(0.002)
Constant	3.144 ***	3.134 ***	3.136 ***	2.594 ***
	(0.008)	(0.007)	(0.014)	(0.137)
Year_fix	yes	yes	yes	yes
City_fix	yes	yes	yes	yes
Observations	4125	4125	4125	4125
R-squared	0.908	0.908	0.908	0.913

Note: *** and ** denote significance at the levels of 1% and 5%, respectively. The numbers in parentheses are standard errors.

reports the estimated results of the impact of digitization levels on carbon emissions. The estimate in column (1) is −0.009 at the level of 1%, the estimates in column (2)–(3) range between −0.015 and −0.013 depending on the set of controls, and the confidence level is 1%, 5%, and 1%. The estimates show that the increase in the level of digitization significantly reduces carbon emissions; according to the estimation coefficient (−0.013) in column (4) and the estimation model, it can be calculated that when the level of digitization increases by 1 unit, the city’s carbon emissions decrease by 1.3% on average.

4.2.2. Robustness Test

This paper conducts robustness tests from the following three aspects: Firstly, due to the estimation bias caused by the measurement error of the indicator, the method of replacing the mainly variable indicator is used to test the robustness. Secondly, because there is a certain endogeneity between the level of digitization and carbon emissions, this paper uses instrumental variables to estimate. Thirdly, considering the spatial correlation of carbon emissions between cities, we use a spatial econometric model to estimate.

Table 7. Robustness test of the impact of digitization levels on carbon emissions: changing variable Indicators.

	(1)	(2)	(3)
VARIABLES	LnCO2_cityI	LnCO2_cityII	LnCO2_cityII
Digital_indexI		−0.013 *
		(0.011)
Digital_indexII	−0.035 **		−0.033 *
	(0.017)		(0.030)
Gdp_per_capita	0.006	0.011 *	0.011 *
	(0.005)	(0.005)	(0.005)
Fdi	−0.304	−0.000	−0.000
	(0.193)	(0.001)	(0.001)
Industry	0.006 ***	0.007	0.007
	(0.002)	(0.004)	(0.004)
Constant	2.595 ***	5.742 ***	5.742 ***
	(0.137)	(0.317)	(0.317)
Year_fix	yes	yes	yes
City_fix	yes	yes	yes
Observations	4125	4095	4095
R-squared	0.913	0.358	0.358

Note: ***, **, and * denote significance at the levels of 1%, 5%, and 10%, respectively. The numbers in parentheses are standard errors. The time window of data in column (1) is 2003–2017, and the time window of data in columns (2)–(3) is 2006–2020.

reports the results of the robustness test when the variable indicators change, CO₂_cityII is used to replace CO₂_cityI, and Digital_indexII is used to replace Digital_indexI. In column (1), LnCO₂_cityI is used as the explained variable, and Digital_indexII is used as the core explanatory variable. The estimate is −0.035 at the level of 5%, LnCO₂_cityII is used as the explained variable in columns (2) and (3), and the estimates are −0.013 and −0.033, respectively, at the level of 10%. The estimated results of changing the variable indicators are very robust.

Due to the impact of endogenous problems, this paper chooses the “smart city” pilot policy and “Broadband China” pilot policy, as a combination of instrumental variables for the digitization level, according to the above research on the impact of the “smart city” pilot policy and the research on broadband China pilot policy (Li et al.) [21]. The regression results of 2SLS in the second stage are shown in

Table 8. Robustness test of the impact of digitization levels on carbon emissions: using instrumental variables.

	(1)	(2)	(3)
	2SLS- Second Stage	2SLS- Second Stage	2SLS- Second Stage
Variables	LnCO2_cityII	LnCO2_cityII	LnCO2_cityII
Digital_indexI	−0.138 ***	−0.198 ***
	(0.029)	(0.041)
Digital_indexII			−0.546 ***
			(0.114)
Gdp_per_capita		0.050 ***	0.053 ***
		(0.009)	(0.010)
Fdi		0.000	0.000
		(0.000)	(0.001)
Industry		0.003 *	0.003 *
		(0.002)	(0.002)
Constant	5.383 ***	4.960 ***	4.955 ***
	(0.039)	(0.118)	(0.119)
Year_fix	yes	yes	yes
City_fix	yes	yes	yes
R-squared	0.231	0.173	0.157
Observations	4095	4095	4095

Note: *** and * denote significance at the levels of 1% and 10% respectively. The numbers in parentheses are standard errors. The time window of data is 2006–2020.

. The estimate in column (1) is −0.138 at the level of 1%; the estimated results with the addition of control variables are shown in columns (2)–(3), and the core explanatory variables in columns (2) and (3) are Digital_indexI and Digital_indexII, respectively. The estimates are −0.198 and −0.546, respectively, at the level of 1%. The results show that the empirical results remain robust when considering endogeneity.

Because there may be a certain spatial correlation in urban carbon emissions, this paper first calculated Moran’s I. The results showed that Moran’s I was 0.378 (p < 0.01), which rejected the original hypothesis of spatial random distribution. Therefore, the spatial econometric model was introduced for the robustness test. The estimation results using the Spatial Durbin Model (SDM), Spatial Error Model (SEM), and Spatial Lag Model (SLM) are shown in columns (1)–(3) of

Table 9. Robustness test of the impact of digitization levels on carbon emissions: using spatial econometric models.

	(1)	(2)	(3)
	SDM	SEM	SLM
Variables	LnCO2_cityI	LnCO2_cityI	LnCO2_cityI
Digital_indexI	−0.008 ***	−0.008 ***	−0.008 ***
	(0.001)	(0.001)	(0.002)
Gdp_per_capita	0.006 ***	0.005 ***	0.004 ***
	(0.001)	(0.001)	(0.001)
Fdi	−0.281 ***	−0.272 ***	−0.300 ***
	(0.078)	(0.075)	(0.081)
Industry	0.005 ***	0.005 ***	0.005 ***
	(0.000)	(0.000)	(0.000)
rho	0.589 ***		0.563 ***
	(0.014)		(0.013)
sigma2_e	0.004 ***	0.005 ***	0.005 ***
	(0.000)	(0.000)	(0.000)
lambda		0.655 ***
		(0.013)
Observations	4125	4125	4125
R-squared	0.050	0.236	0.248

Note: *** denotes significance at the levels of 1%. The numbers in parentheses are standard errors. The time window of data is 2003–2017.

, respectively. The spatial weight matrix used by the model is based on “ Whether the city boundaries are adjacent”; that is: if the city boundaries are adjacent, take 1, otherwise take 0. The estimates in columns (1)–(3) are all −0.008 at the level of 1%, which means that considering the influence of spatial correlation, digitization has a significant negative impact on carbon emissions. In addition, the estimation results show that the carbon emissions of cities have significant positive spatial spillover effects.

4.2.3. Heterogeneity Analysis

Firstly, considering the heterogeneity of the sample group of developed cities and the sample group of underdeveloped cities, the cities are divided into two sample groups for grouping regression according to the per capita GDP of the cities. The estimated results are shown in

Table 10. Group regression (developed cities and underdeveloped cities).

	(1)	(2)
	Developed Group	Underdeveloped Group
VARIABLES	LnCO2_cityI	LnCO2_cityI
Digital_indexI	−0.011 *	−0.029 ***
	(0.007)	(0.010)
Gdp_per_capita	0.007	0.011
	(0.006)	(0.008)
Fdi	−1.321 ***	−0.009
	(0.285)	(0.067)
Industry	0.006 *	0.006 ***
	(0.003)	(0.002)
Constant	2.772 ***	2.463 ***
	(0.305)	(0.137)
Year_fix	yes	yes
City_fix	yes	yes
Observations	2070	2055
R-squared	0.907	0.922

Note: *** and * denote significance at the levels of 1% and 10%, respectively. The numbers in parentheses are standard errors. The time window of data is 2003–2017.

. Columns (1) and (2) are the estimated results of the sample group of developed cities and underdeveloped cities, respectively. The estimated coefficients of the core explanatory variables are −0.011 and −0.029, respectively. From the confidence level, in the sample group of underdeveloped cities, the carbon emission reduction effect of digitalization is more significant.

Secondly, the quantile regression model is used to study the differential impact of digitalization on cities with different carbon emission levels. Figure 7 intuitively shows the estimated results. The trend of digital impact presents a U-shaped change with the rise in carbon emission level: the impact is minimal in the 20th percentile and then increases gradually.

5. Discussion of Influence Mechanisms

5.1. The Mediating Effect of Green Innovation

According to the empirical models (3)–(5), this paper estimates the mediating role of green patent innovation in the process of digital-driven carbon emission reduction based on the mediation effect test model. The estimated results are shown in

Table 11. Test of the mediating effect of green innovation.

	(1)	(2)	(3)	(4)
VARIABLES	Green_patent	LnCO2_cityI	Green_patent	LnCO2_cityII
Green_patent		−0.003 **		−0.003 **
		(0.001)		(0.001)
Digital_indexI	0.919 **	−0.011 *
	(0.450)	(0.006)
Digital_indexII			2.183 *	−0.029 *
			(1.186)	(0.016)
Gdp_per_capita	0.612 ***	0.008	0.646 ***	0.008
	(0.227)	(0.005)	(0.233)	(0.005)
Fdi	4.842 *	−0.290	4.909 *	−0.291
	(2.810)	(0.186)	(2.809)	(0.186)
Industry	−0.020	0.006 ***	−0.021	0.006 ***
	(0.016)	(0.001)	(0.016)	(0.001)
Constant	3.448 **	2.604 ***	3.483**	2.605 ***
	(1.653)	(0.135)	(1.681)	(0.135)
Observations	4125	4125	4125	4125
R-squared	0.363	0.914	0.357	0.914
Number of id	275	275	275	275

Note: ***, **, and * denote significance at the levels of 1%, 5%, and 10%, respectively. The numbers in parentheses are standard errors.

. The estimate in column (1) is 1.634 at the level of 1%, indicating that digitization has significantly promoted regional green patent innovation. In column (2), the estimates of the core explanatory variable (Digital_indexI) and the mediating variable (Green_patent) are −0.011 and −0.003, and the confidence levels are 10% and 5%, respectively. The estimated results of the mediating effect of using Digital_indexII to substitute for Digital_indexI are shown in columns (3) and (4), which are consistent with those in column (1) and column (2). The regression results show that green patent innovation has played an incomplete mediating effect in the process of digitalization promoting regional carbon emission reduction; that is, green patent innovation is one of the paths for digitalization to drive carbon emission reduction in cities, but it is not the only path. Other paths also exist.

5.2. Green Innovation of High-Polluting Listed Companies

This paper studies the green patent innovation of listed companies from 2003 to 2020, focusing on innovative listed companies (excluding companies with 0 patent applications). The empirical model (6) is used to explore the micro-mechanism of the green innovation of high-polluting listed companies. The estimated results are shown in

Table 12. Impact of digitization of cities on green innovation of high-polluting listed companies.

	(1)	(5)	(6)
VARIABLES	Firm_ greenpatent	Firm_ greenpatent	Firm_ greenpatent
Digital_indexI*Pollute_dummy	0.264 **	0.321 ***
	(0.118)	(0.121)
Digital_indexII*Pollute_dummy			0.854 ***
			(0.320)
Pollute_dummy	−6.362 ***	−7.698 ***	−7.705 ***
	(0.652)	(0.833)	(0.833)
Private_dummy		−0.521 ***	−0.522 ***
		(0.180)	(0.180)
Open_time		0.209	0.209
		(0.137)	(0.137)
Manage		−9.083 ***	−9.085 ***
		(1.590)	(1.590)
Profit		11.221 *	11.221 *
		(6.397)	(6.397)
Constant	4.087 ***	1.266	1.274
	(0.606)	(3.224)	(3.225)
Year_dummy	yes	yes	yes
Observations	7117	7117	7117
R-squared	0.010	0.019	0.019

Note: ***, **, and * denote significance at the levels of 1%, 5%, and 10%, respectively. The numbers in parentheses are standard errors. Pollute_dummy is a dummy variable based on whether the listed company belongs to a high-polluting industry. With reference to the “Guidelines for Environmental Information Disclosure of Listed Companies” issued by the Ministry of Environmental Protection of China, the high-polluting industries include thermal power, steel, cement, electrolytic aluminum, coal, metallurgy, chemical industry, petrochemical, building materials, paper, brewing, pharmaceuticals, fermentation, textiles, tanning, and mining.

. The core explanatory variable in column (1) is the cross terms of Digital_indexI and Pollute_dummy, and the estimate is 0.264 at the level of 5%. Then, this paper adds the listed company’s years of establishment (Open_time), net profit of total assets (Profit), management’s shareholding ratio (Manage), whether it is a private enterprise (Private), and other micro-characteristic indicators of listed companies as control variables, and the estimated results are shown in columns (2) and (3). The estimates of the cross terms are 0.321 and 0.854, respectively, and both are significant at 1%. The results show that the increased level of digitization has boosted green innovation in highly polluting listed companies located in the city.

6. Conclusions and Implications

Building a modern governance system for carbon emissions to accelerate the realization of “carbon peaking” and “carbon neutrality” is an issue of global significance and a hot topic of study in academia. Based on the panel data of Chinese cities from 2001 to 2020, this paper examines the promotion effect and impact mechanism of digital governance on carbon emission reduction with smart city pilots as policy impact variables. The main research conclusions are as follows: (1) Digital governance based on smart city construction pilots has significantly reduced the carbon emissions of cities. The estimation results using both DID and PSM-DID models show that the implementation of smart city construction pilots has reduced the carbon emissions of cities by approximately 6.6%, and the effects of this policy will be sustained. The estimation results using alternative indicators or placebos indicate that the empirical conclusions are extremely robust. (2) The digitization level of the city has a significant negative impact on carbon emission, while digitization significantly promotes carbon emission reduction. (3) Green patent innovation of cities has played a significant and incomplete intermediary role in the process of digitization-driven carbon emission reduction; digitization has significantly promoted the green innovation of high-polluting listed companies.

With the development of digital technology, digital governance provides a new way to achieve the “dual carbon” goal. The aforementioned research findings provide the following policy implications: (1) The experience of smart city construction pilots needs to be thoroughly summarized to develop a replicable institutional framework for the digital governance system. This will enhance the government’s digital governance capabilities in the field of environmental governance and compensate for “Market failure”. (2) The government should support the green innovation of enterprises and maximize the technical support function of green patent innovation in the process of carbon emission reduction. For example, the government can support enterprises to increase R&D investment in green technology, and promote the integration of digital technology and green technology. (3) Reducing carbon emissions from high-polluting industries or companies is key to achieving the goal of “carbon peaking and carbon neutrality” and deserves special attention. The government should prioritize supporting these enterprises to develop personalized digital transformation and upgrading plans according to their industrial characteristics and operating models.

Three constraints raised in this paper call for more research. (1) The conclusions of this paper are based on data at the city level, and the smart city construction and digitalization that we focus on are mainly driven by the government, which may be different from the construction of market-oriented smart cities in some countries. Therefore, they should be carefully applied to other nations with weak government governance capabilities. (2) It is necessary to further investigate the effect of digitization on carbon reduction using more industry-level or firm-level data from a more microscopic perspective. Due to the availability of data, the micro study is not enough in this paper. (3) The research conclusions are mainly based on the data from China. Due to the limitations of data acquisition, this paper has not carried out empirical research based on the comparison of domestic and foreign data. All these constraints and deficiencies will be the focus of further research in the future.