Does Public Participation Reduce Regional Carbon Emission?

Abstract

Public participation is playing an increasingly important role in environmental governance. This paper uses panel data from Chinese cities to evaluate the causal relationship between public participation, regional carbon emissions, and regional carbon intensity. We obtain the following conclusions: (1) Public participation significantly reduces regional carbon emissions and regional carbon intensity, which remains robust after a series of robustness and endogeneity discussions. (2) The carbon reduction effect of public participation performs better in eastern regions, regions with higher per capita income, and regions with a concentration of tertiary industries and talents. (3) We divided the public participation into resident participation and environment non-governmental organizations (ENGOs) participation. We found an excellent interactive emission reduction effect with resident participation and a good interaction between resident participation and government and environmental organizations. (4) This paper finds that promoting regional green technology innovation is a significant mechanism for public participation in achieving carbon emission reduction. (5) Finally, this paper found an “inverted U-shaped” non-linear relationship between public participation and regional carbon emissions. The results reveal the importance of public participation in regional carbon emissions and provide an empirical basis for promoting informal environmental regulation.

1. Introduction

Global warming has become widespread worldwide, and climate change has brought general anxiety and increased environmental concerns [1]. Emissions of greenhouse gases, mainly CO₂, are the leading cause of atmospheric warming, and reducing CO₂ emissions is the key to curbing global warming [2]. China has become the world’s largest emitter of CO₂. Mitigating climate warming and achieving low-carbon economic development are important issues for China now and in the future. As a responsible country, China has formulated and implemented a series of emission control policies since the 1970s [3] and has repeatedly committed to reducing CO₂ emissions at international conferences [4]. For example, at the 2015 Paris Climate Conference, China proposed a plan to achieve peak carbon by around 2030. In 2020, the Chinese government reaffirmed its “dual carbon” goal of achieving carbon peak by 2030 and carbon neutrality by 2060.

As an essential tool for solving environmental problems, environmental regulation has been widely used worldwide [5,6]. Environmental regulation can be divided into formal and informal environmental regulation. The effectiveness of formal environmental regulation on carbon emissions has been verified by many scholars [7,8,9,10]. However, it also has some limitations. The government, as the only one responsible for traditional environmental regulation [11,12], has high regulatory costs [13] and has a regulatory deficit problem for small and dispersed polluters [14]. From the perspective of the governance effect, formal environmental regulation depends to some extent on the strength of local law enforcement, and there may be loopholes in lax law enforcement [15]. Therefore, environmental regulation needs the power of the public, and the public’s general concern for the environment can compensate, to some extent, for the government’s deficiency in environmental governance [16]. As direct contactors of the environment, the public usually knows more about some actual environmental conditions than the government, with good interaction effects with government regulations [17].

Due to the limitations of formal environmental regulation, informal environmental regulation represented by public participation has been increasingly recognized by scholars worldwide [18,19,20,21]. In China, an increasing number of public participate in environmental actions in different ways. For example, messages on government websites [22], proposals on environmental issues from Chinese People’s Political Consultative Conference (CPPCC) members [19], participation of ENGOs [23] and comments on online social media platforms [24]. Public participation can be divided into resident participation and ENGO participation according to the form of the involvement. Previous studies have mainly employed environmental organizations as a proxy variable for public participation [23,25], but there are many differences. Compared to the former, ENGOs have a broader horizontal network system to collect the environmental needs of different stakeholders and provide feedback in a centralized manner [26]. Second, compared with individuals, ENGOs can gather environmental power and organize various experts to participate in the research and discussion of environmental issues, making environmental governance more scientific [17]. Therefore, specialized ENGOs can help residents to better participate in environmental issues to a large extent and form a positive interaction between organizations and residents.

Regional carbon emissions, as one of the important components of pollutants, have also been a long-term governance goal of the Chinese government [27]. Many scholars have examined formal environmental regulation from this perspective [4,7,8,9]. This study examines the role of informal environmental regulation of carbon reduction from the public participation perspective. It is significant for China to achieve “peak carbon” and “carbon neutral” goals and extend to developing countries in order to achieve low-carbon development. From the above, we have learned that there are limitations to environmental regulation carried out by formal government. These limitations lead to unsatisfactory emission reduction from formal environmental regulations. However, research still needs to be completed on the carbon reduction effects of informal environmental regulation represented by public participation. Hence, taking regional carbon emissions as a starting point, this paper focuses on the environmental performance of public participation. Also, it attempts to investigate the interactive effects of public participation and government regulation.

The research in this paper achieves the following goals. First, we investigated the causal relationship between public participation and regional carbon emissions using the word frequency search index of “environmental pollution” at the prefectural level as a proxy variable for public participation. The results show that public participation can reduce regional carbon emissions and intensity. This paper performs a series of endogeneity and robustness tests on the findings, including instrumental variables, system GMM (SYS-GMM), model averaging, replace the variable, propensity score matching (PSM), change of the fixed effects and cluster levels, control variables lagged one period, and adding control variables, to obtain reliable experimental results. Second, we discussed the heterogeneity of the effect of emission reduction by public participation from the perspectives of geographic location, income level, talent agglomeration, and industry agglomeration, respectively. Third, we divided public participation into resident and ENGO participation and found positive emission reduction effects between resident participation and formal government environmental regulation and ENGOs participation. Fourth, this paper found that promoting green technology innovation is an effective mechanism for public participation to achieve regional carbon reduction through a mediating effects model. Fifth, we explored the nonlinear effects of public participation in emissions reduction and found that the nonlinear relationship is “inverted U-shaped”.

The marginal contributions of this paper are as follows. First, this paper contributes to the further enrichment of the literature concerning public participation. Previous studies have tended to focus only on the participation of ENGOs [23,25] or resident participation [19,21], rarely placing the two under the same framework for discussion [17]. Drawing on a higher perspective, this paper identifies the positive interactions between the two. Second, we are among the first to attempt to investigate the impact of public participation on carbon emissions. Many scholars have verified the positive environmental contribution of public participation in past studies [19,21,23,25]. However, only some scholars have explored the carbon reduction effects of public participation, and this study fills the gap to some extent. Third, our findings contribute to the ongoing literature on public participation, local government environmental co-governance and the impact of public participation and green technology innovation. We found a positive interaction effect between resident participation and the local government’s formal environmental regulation, and also verified that public participation could promote regional green technology innovation. These findings contribute to inspiring future research. Fourth, we searched for two proper instrumental variables that supplement existing studies and provide new ideas from an empirical perspective. The remainder of this paper is organized as follows. Section 2 reviews the literature in related fields. Section 3 introduces the data and evaluation model used in the study. Section 4 presents the benchmark regression results and a series of empirical results and discusses endogeneity, robustness, and heterogeneity. Section 5 provides further analysis of the results. Finally, Section 6 summarizes the full text and makes relevant policy recommendations.

2. Literature Review

2.1. Formal Environmental Regulation and Carbon Emissions

Previous studies on formal environmental regulation and regional carbon emissions can be summarized into three perspectives: inhibition theory, promotion theory, and complex relationship theory. First, formal environmental regulation reduces regional carbon emissions. Formal environmental regulation has a positive effect on reducing regional carbon emissions through the official nature of government implementation [7,8]. Second, formal environmental regulation increases regional carbon emissions. Due to the differences in socioeconomic development between regions, the policy effects of environmental regulation can be disconnected and may increase regional carbon emissions [28]. Third, there may be a more complex relationship between the two. Some studies have shown an “inverted U-shaped” relationship between formal environmental regulation and regional carbon emissions [29]. When the scope of the study is further narrowed, similar findings are found at the city level [5].

While the carbon reduction effect of formal environmental regulation is controversial, more and more scholars have focused on the shortcomings of formal environmental regulation. In the traditional environmental regulatory system, the government often serves as the only responsible agent [11,12]. Therefore, the government is burdened with higher regulatory costs [13], and there is a regulatory deficit for small and dispersed polluters [14]. In terms of governance effectiveness, formal environmental regulation depends on local enforcement to some extent, leading to the issue of lax enforcement [15]. Therefore, government environmental regulation needs support from the public, and general public concern for the environment can compensate to some extent for the government’s deficiencies in environmental governance [16].

2.2. Environmental Performance of Public Participation

Thanks to the continuous development of information technology, it is now easier than ever for the public to participate in environmental governance. Public attention to environmental issues will help policy makers in their environmental governance efforts [30]. Some studies have shown that public participation can curb pollutant emissions by improving the efficiency of environmental technologies [25]. Moreover, reducing energy consumption is the main way to achieve pollution control, and public participation can improve energy utilization efficiency by reducing total energy consumption [31].

The Internet has irreplaceable advantages over traditional public participation methods due to its transparency, convenience, and extensiveness. It is gradually becoming one of the important tools for supervising environmental issues and influencing the government’s environmental actions to a certain extent [24,32]. According to the media mobilization theory, the media have achieved broad mobilization of the whole of society through public opinion mobilization. Powerful Internet and information and communication technology (ICT) break through the limitations of traditional media and significantly reduce the cost of social mobilization. Rapid information transmission, integration, and dissemination lead to a strong public will, and the public can instantly regulate and correct environmental problems [33]. Compared with more specialized environmental organizations, the public’s attention to environmental conditions through the Internet reflects a more accurate level of public participation to a certain extent.

2.3. Public Participation and Green Technology Innovation

Public participation could promote green technology innovation. According to the Porter hypothesis, appropriate environmental regulation can promote the technological innovation of polluters and thus reduce pollutant emissions. In previous studies of formal environmental regulation, many scholars have repeatedly verified this compensation mechanism [34,35]. As the representative of informal environmental regulation, public participation should be able to transmit this pressure to the supervised subjects and urge them to achieve green technology innovation. Green technology innovation, as an environment-friendly technological advancement, can reduce the environmental pollution caused by it while improving production efficiency. Carbon emissions, regarded as one of the typical pollutants, have been verified by many previous scholars for the carbon reduction effect of green technology innovation [36,37,38]. Therefore, we believe that green technology innovation is an important mechanism for public participation in reducing carbon emissions.

2.4. Non-Linear Performance of Public Participation

Public participation shows a non-linear environmental performance. Public participation will increase the “compliance cost” of the supervised issue, which is likely why the environmental governance role of public participation shows a non-linear relationship. When this environmental threshold cost is low, the supervised subject will not pay attention to the public’s environmental voice, which may lead to further environmental degradation. With the increase in compliance cost, the supervised subject will seek technological innovation in order to reduce cost. Therefore, there may be a non-linear relationship between public participation and environmental performance. This could explain the contradictory findings that environmental regulations promote regional carbon emissions [6] and curb regional carbon emissions [9] in previous works. Also, this nonlinear discussion has been broadly verified in studies related to formal environmental regulation [5,29]. However, there is still a necessity for debate in informal environmental regulation represented by public participation.

3. Research Design

3.1. Model Construction

Referring to previous studies, this paper sets the following two-way fixed effect (TWFE) model as the benchmark regression model [19,39]. The TWFE model controls for the omitted variable problem that varies with individuals but not with time, and varies with time but not with individuals, greatly reducing the endogeneity problem. Although the omitted variable problem still varies with individuals and over time, in subsequent empirical evidence, we used the instrumental variables and SYS-GMM method in subsequent empirical evidence to effectively mitigate this endogeneity.

$$C{E}_{it}={\beta }_0+{\beta }_{1}at{t}_{it}+{\beta }_{2}Contro{l}_{it}+{\delta }_i+{\gamma }_t+{\epsilon }_{it}$$

(1)

$$C{I}_{it}={\theta }_0+{\theta }_{1}at{t}_{it}+{\theta }_{2}Contro{l}_{it}+{\delta }_i+{\gamma }_t+{\epsilon }_{it}$$

(2)

where subscripts $i$ and $t$ indicate the region and year, respectively. The dependent variable $C{E}_{it}$ and $C{I}_{it}$ represent regional carbon emissions and regional carbon intensity. $at{t}_{it}$ is the core explanatory variable, public environmental concern, as a proxy variable for public participation. $Contro{l}_{it}$ represents the control variables that affect carbon emissions and change with $i$ and $t$. ${\delta }_i$ represents the city fixed effect, and ${\gamma }_t$ represents the time fixed effect, which controls the factors that affect all samples. ${\epsilon }_{it}$ is the stochastic disturbance term.

3.2. Variables and Data

3.2.1. Variable Description

Dependent variable and core explanatory variable. The dependent variables of this paper are regional carbon emissions and regional carbon intensity. The former is in millions of tons and the latter in millions of tons per billion, both taken in natural logarithmic form. The core explanatory variable is public participation.

Control variable. In order to avoid omitting variables related to carbon emissions, carbon intensity, and public participation, controlling for them in the regression is necessary. According to existing literature [23,25,40], control variables include: lnindus [41,42], lnFDI [43,44], lnpeople [44,45], lnFis [46,47], lnsci [48], lnGDP [49], lnbus [50] and lnFin [51,52]. These variables reflect the level of social and economic development and pollution emissions. The selection of control variables is based on the literature, mainly considering the relevance of carbon emissions and public participation, in order to reduce the issue of omitted variables as much as possible. Specific definitions are given below.

Lnindus: A region’s carbon emissions are closely related to its industrial structure [36,37]. Referring to the previous work [23], this paper adopts the following formula to measure the local industrial structure. The calculation formula is: 1*(the ratio of the primary industry to GDP) + 2*(the ratio of the secondary industry to GDP) + 3*(the ratio of the tertiary industry to GDP).

lnFDI: The “pollution heaven” hypothesis suggests that pollution-intensive industries will transfer from more heavily regulated regions to weaker regions [43,44]. We believe that regional carbon emissions may also show such a trend of transfer. Therefore, this paper employs the regional ratio of actual FDI use to GDP as a means to incorporate this factor.

lnpeople: More populous regions are potentially more likely to consume more factors of production and living, resulting in more carbon emissions. Therefore, we use the natural logarithm form of the ratio of the regional year-end household population to the regional administrative area as a proxy variable for this factor.

lnFis: The fiscal activities of local governments are closely related to regional carbon emissions, which is determined by the different degrees of local economic development [46,47]. We measure this factor as the natural logarithm of the regional general budget fiscal expenditures ratio to GDP.

lnsci: Technological progress has contributed to increased production efficiency, which may reduce regional carbon emissions [48]. Therefore, we measure it in the natural logarithmic form of the ratio of regional S&T expenditures to GDP.

lnGDP: According to the environment Kuznets curve theory, there is an inverted U-shaped relationship between economic growth and environmental pollution. We believe that there may be some correlation between economic development and regional carbon emissions [49]. Therefore, we introduce the natural logarithmic form of regional gross domestic product as a proxy variable for economic development.

lnbus: The public transport sector in cities, on the one hand, reduces carbon emissions from private cars, and on the other hand, too many public buses bring more carbon emissions. Existing studies suggest that there may be a non-linear relationship [50]. Hence, in our study, this factor is taken into account in the natural logarithm form of the number of public buses operating at the year-end in this paper.

lnFin: Financial development plays an important role in economic growth and technological progress, which could significantly affect regional carbon emissions [51,52]. In this paper, we adopt the natural logarithm form of the ratio of outstanding loans of financial institutions to GDP at the year-end to represent the impact of financial development on regional carbon emissions.

3.2.2. Data Description

This paper uses panel data from 275 cities in China from 2011 to 2019 to assess the impact of public participation on regional carbon emissions and regional carbon intensity, as shown in Figure 1. Previous studies on public participation mostly used ENGOs as a proxy variable [23,25]. ENGOs’ organizational strength is much higher than ordinary residents, which may somewhat cause bias in the research results. In this paper, public participation utilizes the Baidu index, the largest search engine in Chinese, and can reflect to a considerable extent the concern of the general population about environmental issues. The construction process is detailed in Wu et al. [53], and has been covered in previous studies [21,54]. Simply put, the search volume of “environmental pollution” and other related environmental keywords searched on Baidu by the public in the region in one year is counted according to Internet protocol (IP) address. Then, divide it by the past year’s search volume to form the Baidu index. Most of the previous studies that employed the Baidu index as a proxy variable for public participation tended to concentrate only on haze pollution. These studies have constructed variables based on the search volume of haze pollution and related words [53,54]. Some scholars have also combined “haze pollution” and “environmental pollution” to measure public participation [21]. We believe that both of these measures are biased to some extent. The former is limited to haze pollution and cannot fully measure public participation in environmental governance. The latter measure puts “haze pollution” and “environmental pollution” on the same level, which still leads to a biased measurement of public participation. Hence, it is reasonable and reliable to employ only “environmental pollution” and its related words as a proxy variable for public participation in this paper.

We adopt the logarithm of regional carbon emissions as the dependent variable. The carbon emissions were generated by energy consumption, including electricity, coal gas and liquefied petroleum gas, transportation, and heating [55]. We adopted the calculation formula Wu et al. [55] provided, which summed up the carbon emissions of these four aspects and obtained each city’s carbon emissions from 2006 to 2019. Specifically, the carbon emissions of electricity were calculated by the regional baseline emission factors reported by the National Coordination Committee on Climate Change and the electricity consumption provided in the China City Statistical Yearbook. The carbon emissions of coal gas and liquefied petroleum gas were the product of the conversion factors from IPCC 2006 and the consumption of coal gas and liquefied petroleum gas from the China City Statistical Yearbook. The carbon emissions from transportation were the product of the conversion factors from IPCC 2006 and the energy consumption data, which was calculated by the energy consumption of the transportation sectors from the China Statistic Yearbook and the volume of freight traffic and passenger traffic from the China City Statistical Yearbook. The carbon emissions of heating were computed using the data on central heating in cities and the conversion factors from IPCC 2006. Data on central heating in cities were obtained from the China Urban Construction Statistical Yearbook. The government work reports were taken from the official websites of the prefectural governments, manually compiled and analyzed. The green patent data of prefecture-level cities comes from the State Intellectual Property Office. The PITI data was extracted from the PITI annual report on the IPE website. The other variables were obtained from The Statistical Yearbook of Chinese Cities and The Statistical Yearbook of China. Our data are shown in

Table 1. Descriptive statistics.

VARIABLES		Obs.	Mean	S.D.	Min	Max
Dependent variable	CE	2474	1.689	1.097	−2.154	4.603
	CI	2198	−5.690	0.765	−8.397	−2.342
Independent variable	att	2474	2.265	2.440	0	14.04
	att2	2474	11.08	24.72	0	197.0
Control variable	lnindus	2456	5.480	0.0563	5.312	5.626
	lnFDI	2118	−6.365	1.423	−15.25	−3.375
	lnpeople	2462	−2.793	0.911	−6.688	−0.193
	lnFis	2456	−1.823	0.477	−4.584	0.994
	lnsci	2456	12.19	0.980	8.001	15.43
	lnGDP	2459	15.65	1.104	12.45	19.41
	lnbus	2454	6.568	1.075	3.367	10.56
	lnFin	2444	0.174	0.502	−2.123	2.283
	lnEnvExp	1998	10.62	1.503	3.871	15.01
Substitution variable	att_phone	2474	0.948	1.452	0	8.442
	PCE	2473	−4.175	1.143	−7.920	−0.0593
Mediating variable	green	2469	0.353	0.823	0.00100	12.16
Instrumental variable	lnternet	2452	92.03	96.64	1	782
	ER	2464	0.350	0.144	0	1.239
Moderating variable	att_PITI	906	198.2	194.9	0	1069
	PITIscore	906	45.54	16.27	8.300	85.30
	att_gov	2351	9.978	10.72	0	60.68
	lnwscl	2351	4.443	0.197	2.907	4.605

.

4. Empirical Findings

4.1. Benchmark Results

Table 2. Estimation results of the benchmark regression for CE.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	CE_OLS	CE_RE	CE_FE	CE_D-K	CE_BS
att	−0.077 ***	−0.029 **	−0.029 **	−0.029 **	−0.029 **
	(−7.76)	(−2.48)	(−2.24)	(−2.23)	(−2.16)
Control	YES	YES	YES	YES	YES
City FE	NO	NO	YES	YES	YES
Year FE	NO	NO	YES	YES	YES
Observations	2093	2093	2093	2093	2093
R-squared	0.699	0.689	0.641	0.641	0.641

Notes: t/z-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05.

presents the results of the benchmark regression of carbon emissions. Column (1) shows the preliminary regression results, and the core explanatory variables are significantly negative at the 1% statistical level, indicating a significant inhibitory effect of public participation on regional carbon emissions. Column (2) presents the regression results of random panel effects, and the coefficient is reduced but still significant. Column (3) shows the results under the two-way fixed effects model. The core explanatory variables are significantly negative at the 5% statistical level, indicating that public participation can reduce regional carbon emissions. Column (4) shows the results under DK standard errors, and Column (5) shows the regression results for bootstrap sampling 1000 times with the same coefficients and significance as those under two-way fixed effects, demonstrating the robustness of the study’s findings.

The above results illustrate that public participation does reduce regional carbon emissions. We obtain a smaller coefficient than formal environmental regulation’s carbon reduction effect [4,7,9]. Understandably, public participation lacks the organizational rigor and mandatory regulations of government regulations, so it is not as effective as formal environmental regulations in governance. However, public participation has a lower cost and interacts well with formal environmental regulation by the government [17].

Table 3. Estimation results of the benchmark regression for CI.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	CI_OLS	CI_RE	CI_FE	CI_DK	CI_BS
att	−0.170 ***	−0.076 ***	−0.042 ***	−0.042 ***	−0.042 ***
	(−15.24)	(−7.10)	(−3.29)	(−2.62)	(−3.29)
Control	YES	YES	YES	YES	YES
City FE	NO	NO	YES	YES	YES
Year FE	NO	NO	YES	YES	YES
Observations	1865	1865	1865	1865	1865
R-squared	0.206	0.164	0.373	0.373	0.373

Notes: t-values are in parentheses. Significance levels: *** p < 0.01.

presents the results of the benchmark regression of carbon intensity. Column (1) shows the preliminary regression results, and the core explanatory variables are significantly negative at the 1% statistical level, indicating a significant inhibitory effect of public participation on regional carbon intensity. Column (2) presents the regression results of random panel effects, and the coefficient is reduced but still significant. Column (3) shows the results under the two-way fixed effects model. I core explanatory variables are significantly negative at the 1% statistical level, indicating that public participation can reduce regional carbon intensity. Column (4) shows the results under D-K standard errors [56], and Column (5) shows the regression results for Bootstrap sampling 1000 times with the same coefficients and significance as those under two-way fixed effects, demonstrating the robustness of the study’s findings. The results of carbon intensity are close to carbon emissions, but with a larger coefficient. In contrast, public participation shows a more potent inhibitory effect, possibly due to technological development that promotes cleaner production.

4.2. Endogenous Test

4.2.1. Instrumental Variable

The existing research on public participation is primarily measured as proxy variables. Still, due to the errors in the acquisition of various indicators and measurements, it is easy to cause endogenous problems. Additionally, to further reduce the impact of possible omitted variables, this paper adopts Internet penetration and the frequency of “environmental regulation” terms in the work reports of prefecture-level cities as instrumental variables [21,57,58]. Due to its convenience and transparency, the Internet is a burgeoning new media, providing great help for the public to participate in environmental supervision. Theoretically, the higher the popularity of the Internet, the greater the possibility for the public to participate in environmental supervision [21], and the relevance holds. There is no evidence to prove the connection between Internet penetration and regional carbon emissions. The higher the frequency of mention of the term “environmental regulation” in the government work report, the more it shows that the local government pays attention to environmental protection. Hence, the pollution emission level is lower, showing a negative relevance. Similarly, as with word frequency, there is no inevitable connection between regional carbon emissions and carbon intensity, and the homogeneity holds. Considering the visibility of the results, the word frequency data is amplified by 100 times. The regression results are shown in

Table 4. Estimation results of the IV method.

	(1)	(2)	(3)	(4)
	First Stage	Second Stage	First Stage	Second Stage
VARIABLES	att	CE	att	CI
att		−0.0646 **		−0.1236 ***
		(−1.77)		(−2.63)
Internet	0.0051 ***		0.0046 ***
	(4.53)		(4.13)
ER	−0.5422 *** (−3.63)		−0.5326 *** (−3.26)
Control	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	2062	2062	1836	1836
R-squared		0.6384		0.3550
first-stage test statistic of F	17.94		14.38
K-P LM statistic		38.609 ***		30.500 ***
K-Wald F statistic Hansen J statistic		17.945 1.943		14.381 1.060

Notes: t/z-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05.

. The effects of instrumental variables in this paper have passed the exogeneity test of instrumental variables, overidentification test and weak instrumental variable test.

Column (1) and (2) of Table 4 present the regression results for regional carbon emissions. The results show that in the first stage Internet penetration is significantly positive with the core explanatory variables at the 1% statistical level, and word frequency is significantly negative at the 1% level of significance. In the second stage, the core explanatory variable and regional carbon emissions are significantly negative at the statistical level of 5%. In addition, the F values of the first stage and second stages are 17.94 and 17.945, respectively, which are much larger than 10, indicating that the instrumental variable selection passes the weak instrumental variable test. The null hypothesis was accepted in the Hansen test, indicating that the set of instrumental variables was homogeneous. In conclusion, the regressions show that the benchmark regression results remain robust after addressing the endogeneity of the model. Columns (3) and (4) present regional carbon intensity regression results. The analysis is roughly the same as that of regional carbon emissions, proving the robustness of the benchmark regression results. Compared to the coefficients of the interaction terms in Table 2 and Table 3, the absolute value of the coefficients estimated by the two-stage least squares (2SLS) method is greater than the baseline results. This suggests that the actual public participation in the carbon reduction effect is more robust after eliminating the omitted variable problem and measurement bias. Our conclusions remain robust.

In order to further ensure the exogeneity of IV selection, we substituted IV into models (1) and (2) for regression, and the results are shown in Table S1. It can be seen that none of the IVs showed significance. Therefore, the instrumental variables selected in this paper can be considered sufficiently exogenous.

4.2.2. System GMM

Besides IV, we also employed SYS-GMM to further reduce the potential endogeneity problem of the model. The results are shown in

Table 5. Estimation results of the system GMM for CE.

	(1)	(2)	(3)	(4)
VARIABLES	One-Step Normal	Orthogonal	Two-Step Normal	Orthogonal
att	−0.076 ***	−0.056 ***	−0.174 ***	−0.070 ***
	(−3.89)	(−2.65)	(−3.00)	(−2.68)
L.CE	0.578 ***	0.610 ***	0.354 ***	0.552 ***
	(7.22)	(9.52)	(3.11)	(6.20)
Control	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	1849	1849	1849	1849
AR(1)	0.000	0.000	0.000	0.000
AR(2)	0.320	0.456	0.140	0.476
Hansen test	0.165	0.128	0.625	0.132

Notes: t-values are in parentheses. Significance levels: *** p < 0.01. Columns (1) and (2) report the results of one-step and one-step orthogonalization of system GMM. Columns (3) and (4) report the results of two-step and two-step orthogonalization of system GMM.

and

Table 6. Estimation results of the system GMM for CI.

	(1)	(2)	(3)	(4)
VARIABLES	One-Step Normal	Orthogonal	Two-Step Normal	Orthogonal
att	−0.050 **	−0.039 **	−0.039 **	−0.037 **
	(−2.38)	(−2.00)	(−1.99)	(−2.12)
L.CI	0.869 ***	0.911 ***	0.962 ***	0.973 ***
	(4.32)	(13.82)	(6.48)	(17.87)
Control	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	1401	1401	1401	1401
AR(1)	0.002	0.039	0.001	0.000
AR(2)	0.234	0.875	0.133	0.531
Hansen test	0.344	0.354	0.625	0.428

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05. Columns (1) and (2) report the results of one-step and one-step orthogonalization of system GMM. Columns (3) and (4) report the results of two-step and two-step orthogonalization of system GMM.

.

With a preferred lag length of one year, the specification test result for AR (2) indicated no second-order serial autocorrelation. Hansen’s test result also showed no over-identification of the instruments used in the models. Therefore, the models’ fits are found to be reliable for making inferences.

4.3. Robustness Tests

4.3.1. Model Averaging

In order to avoid the unrelated control variables in the model from occupying too high a share in the regression and thus affecting the regression results. We adopted the means of model averaging to give higher weights to the more correlated control variables and thereby ensure the robustness of the regression results. Specifically, model averaging can use various information criteria to solve the possible model uncertainty by giving weights to different indicators [59]. We performed multiple regression again under the screening of AIC, BIC, AICC and NOIC information criteria. bma and wals are commands designd by Magnus et al. [60] to fit a classical linear regression based on the Bayesian estimator and weighted-average least-squares estimator. The results are shown in

Table 7. Model averaging for CE.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	bma	wals	aic	bic	aicc
att	−0.031 ***	−0.063 ***	−0.077 ***	−0.077 ***	−0.077 ***
	(−2.78)	(−6.18)	(−7.46)	(−7.45)	(−7.46)
Control	YES	YES	YES	YES	YES
City FE	NO	NO	NO	NO	NO
Year FE	NO	NO	NO	NO	NO
Observations	2093	2093	2093	2093	2093

Notes: t-values are in parentheses. Significance levels: *** p < 0.01.

and

Table 8. Model averaging for CI.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	bma	wals	aic	bic	aicc	noic
att	−0.168 ***	−0.157 ***	−0.170 ***	−0.168 ***	−0.170 ***	−0.107 *
	(−12.90)	(−13.25)	(−13.90)	(−12.90)	(−13.90)	(−1.60)
Control	YES	YES	YES	YES	YES	YES
City FE	NO	NO	NO	NO	NO	NO
Year FE	NO	NO	NO	NO	NO	NO
Observations	1865	1865	1865	1865	1865	1865

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, * p < 0.1.

.

After assigning higher weights to the relevant control variables according to different information criteria, the results remain significantly negative, indicating that the results above are robust.

The regressions of carbon intensity also showed no significant changes in the results after model averaging, again demonstrating the robustness of the results above.

4.3.2. Exclusion of Other Interference Policies

Given that there are still some environmental policies for carbon emissions and carbon intensity in the same period, these policies are likely to have a corresponding impact on the dependent variables, thus causing errors in the regression results of this paper. The environmental policies and interference reports excluded in this study mainly include the carbon trading market pilot policy and the PITI report issued by IPE. The former has influenced this study primarily in the carbon trading markets in Shenzhen and Guangdong Province since 2013, and in Hubei Province and Fujian Province since 2014. The latter influences this paper by including 108 cities since 2011 (including Dongguan, Zhongshan, Linfen, and other cities) and seven cities since 2013 (Zhenjiang, Sanmenxia, Zigong, Deyang, Nanchong, Yuxi, and Weinan). We added two policy dummy variables to the model, and the results are shown in

Table 9. Estimation results of excluding other interference policies and replacing the variable.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	CE	CI	CE	CI	PCE
att	−0.029 **	−0.043 ***			−0.037 **
	(−2.20)	(−3.42)			(−2.44)
att_phone			−0.028 **	−0.042 ***
			(−2.10)	(−3.27)
PITI	−0.064	−0.142
	(−0.46)	(−1.15)
ETS	0.012	−0.035
	(0.20)	(−0.62)
Control	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	2093	1865	2093	1865	2093
R-squared	0.641	0.374	0.641	0.374	0.620

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05.

.

According to the regression results in Table 9, Column (1) is the regression on carbon emissions, and its coefficient and significance remain unchanged and significantly negative. Column (2) shows the regression of carbon intensity, and the coefficient has increased. The above results indicate that after excluding other interfering policies, public participation still has a significant inhibitory effect on regional carbon emissions and regional carbon intensity, and other interfering policies do not trigger such carbon reduction effects.

4.3.3. Replace the Variable

The mobile phone has become the most widely used Internet terminal due to its convenience. In order to reduce the bias brought by the specific accounting method of public environmental concern, this paper adopts the Baidu index of the mobile phone as the core explanatory variable substituted into Model (1) and (2) for regression, and the results are shown in Columns (3) and (4) of Table 9. There are differences in economic development and industrial structure among different cities, which will affect the regression results of total carbon emissions and carbon intensity to a certain extent and may bring some regression bias. Therefore, this study utilizes carbon emissions per capita as a proxy for the dependent variable, and the results are shown in Column (5) of Table 9.

The results in Column (3) and (4) of Table 9 show that the results are still significantly negative, and the coefficient size remains the same after the core explanatory variables are replaced with the public environmental concern on mobile phones. The results in Column (5) show that the regression of per capita carbon emissions is still significantly negative, and its coefficient is between the total carbon emissions and carbon intensity, which proves that public participation has a significant inhibitory effect on carbon emissions in three different aspects: quantity, intensity, and per capita.

4.3.4. Propensity Score Matching

Considering the differences in individual characteristics between sample cities, such as openness, scientific and technological level, etc., may lead to different experimental results. In this section of the paper, the whole sample is divided into experimental and control groups based on the median values of the core explanatory variables. InFDI and Insci are taken as characteristic variables, and the nearest neighbor matching method (1:4), caliper matching method and nuclear matching method are used, respectively, to replace the matched samples into Models (1) and (2) for regression. The results are shown in

Table 10. Estimation results of propensity score matching.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	CE_neigh	CI_neigh	CE_radius	CI_radius	CE_kernel	CI_kernel
att	−0.029 **	−0.038 ***	−0.030 **	−0.043 ***	−0.030 **	−0.043 ***
	(−2.13)	(−2.97)	(−2.29)	(−3.37)	(−2.29)	(−3.39)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	1932	1689	2073	1847	2073	1848
R-squared	0.644	0.381	0.643	0.373	0.643	0.373

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05.

. The comparison between samples before and after applying nearest neighbor matching, caliper matching and kernel matching are provided in Tables S2–S4, respectively.

Columns (1), (3) and (5) of Table 10 present the regression results for nearest neighbor matching, caliper matching, and kernel matching for regional carbon emissions, respectively. Compared to the results in Column (3) of Table 3, neither the sign nor the coefficients changed significantly, indicating the robustness of the benchmark results in this paper. Similarly, Columns (2), (4) and (6) of Table 10 presents the results for regional carbon intensity, and again the sign remains the same and the coefficients do not change significantly, proving the robustness of the benchmark results above.

4.3.5. Change of the Fixed Effects and Cluster Levels

This paper refers to Chen and Xie [61] to mitigate the effects of unmeasurable factors in other dimensions. This section retains city and year fixed effects and city-level clustering. Province, province–years interaction, and province and province–years interaction fixed effects are added separately, and province, city–years interaction, province–city interaction, and province–years interaction standard errors are added separately for clustering. The regression results of regional carbon emissions and regional carbon intensity are shown in Figure 2.

The blue line segment in Figure 2 represents the 95% confidence interval, the orange point in the middle represents the regression coefficient, and the vertical red line represents the baseline 0 value. The confidence intervals of the core explanatory variables of the above fourteen regression sets are different from the baseline 0 value, so it can be assumed that the results above remain robust.

4.3.6. Control Variables Lagged One Period

Consider that GDP is used in the variable construction of CI, while GDP and its related indicators are still selected in our control variable selection. This may lead to severe collinearity and endogeneity. Hence, we adopted the strategy of control variables lagged one period to deal with this problem. The results are shown in Table S5.

Our core explanatory variables remain significant after we lag the control variables for one period, indicating that the conclusions of this paper still hold.

4.3.7. Adding Control Variables

Although the application of IV compensates for the omitted variables to some extent, this paper adds control variables in order to enhance the robustness of the results. This paper adds environmental protection expenditures to the benchmark regression, measured in the natural logarithm form of environmental protection expenditures in prefecture-level municipalities. Data are available from the regional statistical yearbooks. The results are shown in Table S6.

The coefficients and significance of the core explanatory variables stayed mostly the same after the adding of lnEnvExp. This indicates that our conclusions remain robust.

4.4. Heterogeneity Test

4.4.1. Subsample Regression by Geographic Location

Given China’s vast territory, cities in different geographical locations vary significantly in terms of economic structure and policy implementation, which may result in different levels of effectiveness of public participation. According to the division regulations of the 5th session of the 8th National People’s Congress, this paper divides the sample into Eastern, Central and Western regions. The results are shown in

Table 11. Heterogeneity of geographic location test.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	CE_E	CI_E	CE_C	CI_C	CE_W	CI_W
att	−0.045 **	−0.065 ***	−0.004	−0.007	−0.054	−0.069 *
	(−2.39)	(−3.55)	(−0.18)	(−0.33)	(−1.62)	(−1.88)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	839	746	935	832	319	287
R-squared	0.644	0.365	0.692	0.464	0.622	0.384

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

.

Columns (1) and (2) of Table 11 present the regression results of regional carbon emissions and regional carbon intensity from the sub-sample in the eastern region, and the coefficients are significantly negative. Compared with the whole-sample regression, its coefficient is larger, indicating that public participation has a more significant inhibitory effect on regional carbon emissions and regional carbon intensity in the eastern region. Columns (3) and (4) present the regression results of the central region, and Columns (5) and (6) present the regression results of the western region. Column (6) is significantly negative at the 10% statistical level, indicating that public participation can inhibit regional carbon intensity to some extent in the western subsample. At the same time, the rest of the results are not statistically significant. One possible explanation is that the eastern region has a higher level of economic development than the central and western regions, and there is a large agglomeration of industries. In order to create a better business environment, the eastern region pays more attention to public feedback on environmental issues. The more concentrated industrial agglomeration facilitates the government to manage pollution problems. Thus, the inhibitory effect of public participation on regional carbon emissions and regional carbon intensity is more pronounced in the eastern region sample, which is also consistent with previous studies [21,22].

4.4.2. Subsample Regression by Industrial Agglomeration

There is a significant imbalance in industrial development among regions, and the varying emphasis on industrial development will determine whether the government adopts flexible policies on pollution to a certain extent. Cities with a high concentration of high-emission industries, where the government depends on these industries for tax revenue, will adopt a more flexible attitude towards the polluting behavior of these enterprises, choosing to ignore the carbon emissions and carbon intensity of the region in the trade-off between economic development and environmental protection [62]. In this paper, the whole sample is divided into five equal parts according to the proportion of the tertiary industry in GDP, and 1, 3 and 5 are taken as the low, medium and high subsamples for regression, respectively (exclude 2, 4). The results are shown in

Table 12. Heterogeneity of industrial agglomeration test.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	CE_H	CI_H	CE_M	CI_M	CE_L	CI_L
att	−0.080 ***	−0.080 ***	0.019	−0.019	0.042	0.071
	(−3.58)	(−3.52)	(0.43)	(−0.35)	(1.02)	(1.18)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	387	346	443	394	423	378
R-squared	0.588	0.320	0.695	0.482	0.641	0.422

Notes: t-values are in parentheses. Significance levels: *** p < 0.01.

.

Column (1) and (2) of Table 12 respectively present the regression results of public participation in regional carbon emissions and regional carbon intensity in the subsample with a high degree of tertiary industry agglomeration. The regression coefficients are significantly negative at the statistical level of 1%, which is far greater than the benchmark regression, indicating that in the subsample with a high degree of tertiary industry agglomeration, public participation can achieve better performance in inhibiting regional carbon emissions and regional carbon intensity. Columns (3) and (4) and Columns (5) and (6) present the regression results of subsamples with a medium and low degree of concentration, respectively. It can be seen that they are not statistically significant, and that the governance effectiveness of public participation needs to be improved.

4.4.3. Subsample Regression by Income Level

According to the environmental Kuznets curve, when people’s income levels reach a certain stage, the public will become increasingly concerned about the environmental friendliness of their living environment. Under this view, the effect of public participation will be better in high-income areas than in less-developed areas. In this paper, the whole sample is divided into five equal parts according to per capita income level, and 1, 3 and 5 are taken as the low, middle and high subsamples for regression (exclude 2, 4). The results are shown in

Table 13. Heterogeneity of income level structure test.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	CE_H	CI_H	CE_M	CI_M	CE_L	CI_L
att	−0.046 ***	−0.057 ***	0.030	0.017	−0.061	−0.087
	(−2.78)	(−2.81)	(0.97)	(0.50)	(−1.02)	(−1.20)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	447	400	446	397	346	310
R-squared	0.646	0.0357	0.698	0.503	0.706	0.466

Notes: t-values are in parentheses. Significance levels: *** p < 0.01.

.

Columns (1) and (2) of Table 11 report the results for regional carbon emissions and regional carbon intensity for the subsample of high-income regions. The results are significantly negative at the 1% statistical level, indicating that the governance effect of public participation performs better in high-income areas, which is consistent with previous studies [63].

4.4.4. Subsample Regression by Talent Agglomeration

Existing studies generally agree that there is a strong positive correlation between educational level and environmental behaviors and intentions [64]. Therefore, when a region has many highly educated talents, such talent agglomeration advantage will spontaneously form a positive mechanism of environmental supervision and promote regional environmental protection. In this paper, the whole sample is divided into five equal parts according to regional education expenditure, and 1, 3 and 5 are taken as low, medium and high subsamples for regression (exclude 2, 4). The results are shown in

Table 14. Heterogeneity of talent agglomeration test.

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	CE_H	CI_H	CE_M	CI_M	CE_L	CI_L
att	−0.088 ***	−0.092 ***	−0.010	−0.023	0.056	0.023
	(−4.03)	(−4.14)	(−0.17)	(−0.36)	(0.60)	(0.22)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	455	407	408	363	371	329
R-squared	0.728	0.464	0.814	0.614	0.471	0.242

Notes: t-values are in parentheses. Significance levels: *** p < 0.01.

.

Columns (1) and (2) of Table 14 report the results for the high-talent agglomeration level subsample, and the regression results are all significantly negative at the 1% statistical level, indicating that public participation is more effective in governing regional carbon emissions and regional carbon intensity in regions with a high talent agglomeration. However, this more significant inhibition does not exist in the subsample with a lower level of talent accumulation in a statistical sense.

5. Further Analysis

5.1. Interaction with Government and ENGOs

5.1.1. ENGOs

Previous studies often used ENGOs as proxy variables for public participation [23,25]. However, ENGOs are significantly different from ordinary residents in terms of organizational strength, relationship network and financial energy, leading to differences in how polluters deal with the environmental problems called for solutions. Therefore, this study subdivides public participation into resident and ENGOs participation [17]. The data of ENGOs participation adopts the 2011–2018 PITI score of prefecture-level cities released by IPE, which has been widely used in previous studies [65]. In this paper, the interaction terms between PITI score and public environmental attention are constructed and substituted into regression Models (1) and (2). The results are shown in Columns (1) and (2) of Table 13.

It can be seen that the interaction between the PITI score and public environmental concern is significantly negative, indicating that there is good interaction between resident participation and ENGO participation, which can effectively reduce regional carbon emissions and regional carbon intensity. Resident participation can help ENGOs better reflect the public’s environmental issues at a more grass-roots level.

5.1.2. Government

The government has long been seen as the only one responsible for environmental issues [11], and environmental power from the government has certain limitations such as high costs [13] and lack of supervision [14]. The effectiveness depends on law enforcement efforts [15]. As a spontaneous environmental protection behavior of the public, public participation can urge polluters to control emissions with low-cost supervision behaviors and supplement government regulation to a certain extent [16]. In this paper, the logarithm of centralized sewage treatment rate is taken as the proxy variable of government environmental regulation [66], and the interaction terms with public environmental attention are constructed and substituted into the regression Models (1) and (2). The results in Columns (3) and (4) of

Table 15. Interaction with government and ENGOs, mechanism analysis and nonlinear discussion.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
VARIABLES	CE	CI	CE	CI	green	CE	CI
att_PITI	−0.001 *	−0.001 **
	(−1.85)	(−2.57)
att_gov			−0.111 *	−0.143 **
			(−1.68)	(−2.03)
att2						−0.011 ***	−0.008 ***
						(−4.92)	(−3.31)
att	0.028	0.022	0.482	0.616 *	0.235 ***	0.097 ***	0.053
	(1.12)	(1.02)	(1.62)	(1.93)	(4.74)	(3.29)	(1.58)
PITIscore	0.004 **	0.004 **
	(2.30)	(2.32)
lnwscl			0.206	0.185
			(1.56)	(1.26)
Control	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES
Observations	809	709	2005	1793	2091	2093	1865
R-squared	0.609	0.341	0.657	0.400	0.371	0.648	0.379

Notes: t-values are in parentheses. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.

.

It can be found that the interaction term between government regulation and public environmental concern is significantly negative, indicating that there is good interaction between public participation and government regulation. Such interactive relationship can effectively reduce regional carbon emissions and regional carbon intensity, and public participation does play a role in supplementing government environmental regulation to a certain extent [17]. Moreover, it can be found that the interaction term coefficients of government regulation are much more significant than those of ENGOs, and it is well understood that even ENGOs’ participation with its stronger organization also lacks the organizational rigor and regulatory compulsion of official endorsement, and cannot be close to government environmental regulation at the level of governance effect.

5.2. Mechanism Analysis

Environmental regulations will impose certain environmental thresholds on polluters, and cost pressures will force them to seek green technologies and innovations to offset economic losses [34,67]. This paper selects the sum of green patent applications in prefecture-level cities as the proxy variable of green technology innovation. Considering the visualization of results, the variables are reduced by 1000 times, and the regression results are shown in Column (5) of Table 15.

From the regression results in the above table, it is clear that there is a positive contribution of public participation to green technology innovation at the 1% statistical level. Previous studies have confirmed that green technology innovation significantly inhibits regional carbon emissions and intensity [36,37,38]. Therefore, this paper believes that promoting regional green technology innovation is a mechanism for public participation in reducing regional carbon emissions and regional carbon intensity.

5.3. Nonlinear Relation Analysis

Considering that there is a dynamic relationship between the compliance costs of public participation and the effect of environmental governance, this paper suggests that there may be a non-linear relationship between public participation and regional carbon emissions and regional carbon intensity. Such a non-linear relationship has been confirmed in previous studies of formal environmental regulation on regional carbon emissions [5,29]. In this paper, a quadratic term of the core explanatory variables was constructed to be added to the regression, and the results are shown in Columns (6) and (7) of Table 15.

From the regressions, we find that the quadratic terms of the regressions for regional carbon emissions and regional carbon intensity are significantly negative at the 1% level, thus showing an “inverted U-shaped” relationship. Moreover, we also apply the “utest” command to obtain the extreme point and intervals of the variables, see Table S7. The extreme point of CE and CI fall at 4.615594 and 3.362124, respectively, which are included in the interval of att ([0; 14.03644]). Therefore, we think this “inverted U-shaped” nonlinear relationship exists. This may help us understand why previous works related to environmental regulation have reached contradictory conclusions [6,9]. Furthermore, such pressure for supervision needs to be strengthened when the public engages in environmental issues to avoid counter-productive effects.

6. Conclusions and Limitations

Does public participation reduce regional carbon emissions and regional carbon intensity? Answering this question is of great significance for China to implement the goals of carbon peaking and carbon neutrality, and then to extend this experience to global green governance. Air pollution is a matter of public health and welfare [68,69]. To capture the carbon reduction effect of public participation, we use the search index of “environmental pollution” on the Baidu platform as a proxy variable to explore the causal relationship between public participation and regional carbon emissions and intensity. We employ the TWFE model to estimate the impact of public participation on regional carbon emissions for 275 cities in China from 2011–2019. We then analyzed the heterogeneous effects of public participation on carbon reduction effects based on geographic location, income level, talent agglomeration, and industry agglomeration. Furthermore, we divided public participation into resident participation and ENGO participation and explored the interaction between resident participation with formal environmental regulation and ENGOs’ participation. With green technology innovation as a mediating variable, this paper explores the mechanism of public participation in regional carbon emissions. Finally, we tried to verify the potential non-linear relationship between public participation and regional carbon emissions and intensity. The main findings are as follows:

First, public participation can significantly reduce regional carbon emissions and intensity. On the one hand, general environmental supervision forces polluters to reduce emissions and promotes environmental efficiency [23,70]. On the other hand, the public is more aware of the natural environmental conditions, which could complement some of the limitations of the government in environmental governance [17,22].

Second, public participation achieves better carbon reduction effects in eastern regions with high income, talent agglomeration, and tertiary industry agglomeration. These regions have better economic performance, population quality, and business environment, which we suggest as a possible reason why public participation shows better carbon reduction effects [21,63,64].

Third, a positive interactive carbon reduction effect exists between resident participation with formal environmental regulation and ENGOs’ participation. Previous studies have rarely further divided public participation. We analyzed resident participation and ENGO participation under the same framework in order to validate the possibility of such collaborative carbon reduction governance empirically.

Fourth, promoting green technology innovation is an important mechanism for public participation in reducing regional carbon emissions and intensity. This paper finds a significant positive relationship between public participation and regional green technology innovation. One possible explanation is that public participation raises the production costs of polluters, who are financially motivated to promote technological innovation [36]. Many previous studies have verified the carbon reduction effect of green technology innovation [36,37,38].

Fifth, there is an inverted U-shaped non-linear relationship between public participation and regional carbon emissions and intensity. Public participation raises the compliance cost for polluters. When this cost is lower than the additional cost required for low-carbon production, polluters will continue their original production ways. As the compliance cost increases, polluters will shift to low-carbon production, thus forming an inverted U relationship [5,29].

Regarding the research design of this paper. We set the TWFE model as the benchmark regression model, which effectively circumvents the endogeneity to a certain extent. Further, we selected IV and SYS-GMM to mitigate the endogeneity problem. We employ model averaging, replace the variable, control variables lagged one period, and add control variables in order to reduce the influence of specific control variables on the study findings. Meanwhile, we obtained more robust conclusions using the exclusion of other interference policies, PSM and change of the fixed effects and cluster levels.

Overall, this paper is generally consistent with previous literature on the effect of formal environmental regulation on regional carbon emissions [7,8]. Our results play a crucial role in the theory and contribute to practice. It provides convincing evidence that public participation can effectively reduce regional carbon emissions and intensity. Although this paper focuses on Chinese cities as the main subject of the study, its findings can be extended to a wide range of developing countries currently facing development and environmental protection dilemmas. Specifically, this paper proposes the following policy recommendations: (1) Further smooth public information feedback channels, strengthen public awareness of environmental protection and health knowledge, vigorously promote public participation in environmental supervision, and play an important role in public environmental protection. (2) Public participation should be promoted in areas where there is a willingness to create a better business environment, tertiary industry agglomeration, higher income levels, and talent agglomeration, and where the environmental performance of public participation is better. (3) Support the development of ENGOs and broaden their information transmission channels. The government should protect ENGOs’ establishment and operation through legislation, provide necessary financial and policy support, and bring into play the good interaction between ENGOs’ participation and residents’ participation. (4) Improve the government environmental supervision laws and regulations, while improving the intensity and efficiency of government supervision, encouraging a good interaction between government regulation and public participation. (5) While promoting public participation in order to reduce regional carbon emissions and regional carbon intensity, we should focus on the level of development of green technology innovation for polluters. (6) Deepen and strengthen the intensity and breadth of public participation and avoid the inverted U-shaped relationship between public participation and regional carbon emissions and regional carbon intensity, which will lead to counterproductive governance effects.

Some limitations in this paper may inspire future research. (1) Due to data limitations, the public participation data started in 2011, and the carbon emission data ended in 2019, so the experimental results in this paper are based on 2011–2019, and future studies can be conducted based on better data. (2) Some data in the China Urban Statistical Yearbook need to be included. For the robustness of the results, this paper does not use interpolation to supplement it but instead uses unbalanced panel regression. (3) The public participation data in this paper refer to previous scholars’ practices. Future public opinion analysis based on machine learning methods can be used for deeper sentiment learning to obtain more robust experimental results.