Is More Always Better? Government Attention and Environmental Governance Efficiency: Empirical Evidence from China

Abstract

In recent years, the thorough implementation of China’s green development concept has compelled local governments to devote more attention to environmental issues. This study aimed to verify whether increased government environmental attention (GEA) can sustainably ensure the implementation of environmental governance, particularly air pollution control. Using government work reports (GWRs) from local governments, this study employed machine learning methods to identify and quantify the attitudes of government officials as expressed in policy texts. A weighted dictionary method was used to quantify GEA from 2011 to 2016. The results of spatial econometric models indicated that air pollution exhibited positive spatial clustering effects across different regions, with the Yangtze River Delta and the Beijing–Tianjin–Hebei region being classified as high–high areas, while the western regions were classified as low–low areas. Baseline regression results showed that increased GEA can improve the effectiveness of pollution control, but excessive attention leads to a decline in governance efficiency. Overall, this study helps explain the unsustainability of campaign-style environmental governance and provides guidance for local governments on the rational allocation of attention when addressing environmental issues.

1. Introduction

The escalating global environmental crisis has spurred discussions on the effectiveness of various government interventions in mitigating environmental degradation and enhancing environmental performance. Air pollution is an important aspect of environmental governance, as air pollution reflects the effectiveness of environmental governance [1]. In 2013, China launched a top-down environmental campaign called the “Air Pollution Prevention and Control Action Plan”, which achieved significant results in air pollution control within a short period through a campaign-style governance approach. Campaign-style governance is a policy implementation approach employed by the dominant political party or the central government to address unexpected events or major domestic social issues [2]. Through this form of governance, the requirements and goals of the central government are transmitted to the local government, and then the attention allocation structure of the local government changes. This method has played an important role in the field of environmental governance through the mobilization of administrative resources [3].

The effectiveness of campaign-style environmental governance is a central focus of debate within the relevant literature. Specifically, traditional views tend to regard campaign-style environmental governance as policy interventions that are most likely to produce short-term policy effects [4]. However, the recent literature suggests that campaign-style enforcement can actually generate long-term, sustained positive policy outcomes through the use of strong external stimuli and a re-coupling mechanism [5,6]. While the existing literature provides valuable insights into campaign-style environmental governance from a policy evaluation perspective, there is no consensus on its governance efficiency, necessitating further academic exploration. For example, in the context of campaign-style environmental enforcement initiated by the central government, why do different local governments adopt various policy response measures, leading to differences in governance outcomes? How can reliable statistical data be used to better demonstrate the effectiveness of environmental governance campaigns?

Government attention is a crucial administrative resource that directly reflects local governments’ responses to the central government’s campaign-style governance. It also serves as a prerequisite for local governments to formulate policies and allocate resources [7]. This provides us with a unique opportunity to review and supplement the literature on the efficiency of environmental governance from the source of policy output. Existing research has preliminarily verified the role of government attention in environmental governance. The impact of government attention on climate change was analyzed using presidential speech documents from the Mexican government website in 2020 [8]. In another study, the potential spatial spillover effects of air pollution were considered by incorporating spatial factors and constructing regression equations to analyze the spatial characteristics and dynamic changes in environmental governance in 2022 [9]. They used the Spatial Durbin Model (SDM) to analyze the impact of GEA on air pollution. Other scholars have analyzed the impact of government attention on carbon and sulfur dioxide emissions [10,11,12]. Without exception, these studies were based on the assumption of the complete rationality of government officials and have verified that a sustained shift of GEA significantly improves environmental governance efficiency. However, in reality, government officials are boundedly rational and cannot process external information equally; under the intervention of higher-level pressure, local governments often devote more attention to specific issues, crowding out attention to other issues and increasing the pressure on local governments’ resource endowments. Under these conditions, whether substantial GEA can continue to improve the level of environmental governance requires further study.

Another key issue is how to scientifically measure GEA and accurately assess the administrative resources allocated by the government to environmental issues. One aspect of this is the selection of policy documents. Different countries and regions exhibit variations in the content, objectives, and structure in their public policies [13], which also determines the diversity of policy texts to measure government attention, such as speeches, documents, reports, and congressional hearing materials [14,15,16,17]. In China, the GWRs of local governments are formulated by comprehensively considering higher-level policy directives, local resource endowments, and development prospects. This means that these reports are annual plans that are developed based on a thorough assessment of local realities, and they are reliable, authentic, and continuous. The content of these reports includes past achievements, the next year’s goals, and specific measures, serving as authoritative documents guiding grassroots work. Additionally, the annual local government work reports are published and archived on their official websites, ensuring public supervision, enhancing government governance efficiency, and fully realizing the functions of local governments.

Regarded as an expression of governmental consciousness, GWRs are commonly referred to as the government’s “baton” for coordinating resource allocation and practical actions [18]. They plays a central role in conveying policy signals to functional government departments, businesses, and the public, making them useful for analyzing China’s government attention [19].

The other aspect is the choice of measurement methods. Many scholars employ the dictionary method, utilizing keyword frequency ratios to measure government attention [20]. This approach overlooks differences among the semantic intensities of similar keywords, thereby making it impossible to discern variations in different local governments’ prioritization of the same initiatives. This limitation results in high homogeneity of attention indicators and reduced accuracy. Verbs, adjectives, adverbs, and phrases in policy texts can be used to analyze attitudinal differences in governments. For example, the phrases “promote”, “greatly promote”, and “strictly control” show different semantic strength, demonstrating a trend from weak to strong. For this reason, we integrated and extended words describing the semantic intensity of policy text using machine learning. This approach constructed an intensity index for government attention to enable a more precise elucidation of government attention disparities and a more accurate validation of the impact of government attention.

In order to verify the relationship between GEA and environmental governance efficiency, we first used 4470 GWRs to construct a dictionary reflecting the semantic strength of the words used in policy texts with the help of machine learning methods, and we then measured the GEA index by weighting the environmental keywords. Secondly, we conducted a series of spatial autocorrelation tests to verify the spatial spillover effect of air pollution and selected SDM as a reasonable spatial econometric model. Thirdly, we verified the nonlinear relationship between GEA and air pollution.

Our research contributes to the existing literature in the following ways. Firstly, the semantic strength of policy texts was considered in the process of measuring government attention. The existing research only focuses on the frequency of environmental keywords or the proportion of the word frequency in the text, making it impossible to determine how each government’s efforts differ in the same affairs. We combined manual coding with machine learning to build a degree word dictionary that reflects the semantic strength of words used in policy texts, thereby further improving the quantitative method of government attention. Secondly, we identified the nonlinear relationship between GEA and the efficiency of environmental governance in the context of centralized environmental improvement, which provided an explanation for symbolic implementation in practice from the perspective of government attention. In addition, we also controlled for the variables that may affect air pollution by transforming provincial variables into city-level, which provided a reference for subsequent research.

2. Theory and Hypothesis Development

2.1. Theoretical Setting

The theoretical framework adopted in this paper combines the institutional background of the pressure-driven system and the attention-based view. Campaign-style environmental governance originates from the pressure-driven system, which mobilizes substantial resources and forces in the short term through top-down specific goal transmission, performance assessments, and official promotions, to achieve rapid development and policy implementation [21,22]. The attention-based theory suggests that organizational decisions and behaviors are not only the result of decision makers’ attention allocation but also of external factors influencing this allocation process. Therefore, organizational decisions are closely related to the institutional environment in which decision makers operate [23]. Managers’ attention allocation varies in response to the external institutional environment, leading to different organizational behaviors and governance outcomes [24]. Of course, the institutional logic of organizations is not static; it changes with key external events and the evolving environment [25]. This phenomenon provides a theoretical basis for analyzing the relationship between government attention and environmental governance outcomes in the context of campaign-style environmental governance, suggesting that such governance can influence environmental outcomes by affecting government attention allocation.

2.2. Hypothesis Development

This study proposed the following hypothesis:

Hypothesis 1.

The GEA has a U-shaped impact on air pollution: as the GEA increases, air pollution levels decrease; however, when the GEA gets too high, air pollution levels increase alongside the government attention.

In the following sections, we provide detailed arguments to support this hypothesis.

2.2.1. Problem Orientation: The Positive Impact of Increased GEA on Environmental Governance

Government attention is a crucial resource serving as a precursor to policy outputs and the allocation of various resources [7]. Faced with restrictive policy directives, local government officials often mobilize limited attention resources, concentrating them to achieve policy goals, thereby mitigating risks and gaining political capital [26]. Under the pressure-driven system, higher-level governments delegate administrative goals to lower-level governments, often escalating the pressure at each tier. Through various administrative measures, lower-level governments coordinate with functional departments to take collective action, ensuring they successfully meet or even exceed the assigned evaluation tasks [27].

Increased GEA improves the probability that policymakers will identify potential problems and make effective decisions. However, this conclusion assumes that attention is properly allocated. Based on the bounded rationality theory, optimal attention allocation must align with an organization’s own resource endowment and developmental needs [28]. Once local governments exceed their actual capacities, leaders may cease allocating the limited time and resources to policy implementation. Instead, they may prioritize completing paperwork that circumvents the objectives of policy execution and instead solely pursues formal compliance [29]. This leads to a results-oriented decision-making process, as opposed to a question-oriented one. An excessive allocation of attention alters the decision-making process, which is in line with the decision-making model employed by rational people [30] but does not ensure the correctness of the decision making.

2.2.2. Result Orientation: The Negative Impact of Increased GEA on Environmental Governance

Attention is a limited resource, and when managers focus on environmental governance, they must reduce the time they spend on other issues, leading to an increase in the opportunity cost of environmental governance. Moreover, attention does not have market liquidity, so an authorization for increased attention on a specific issue from the higher government cannot increase the local government’s upper limit of attention [31]. Government attention cannot generate value without the effective allocation of other resources. Even if attention continues to increase, there is no guarantee that the resources necessary to support this increased attention will be adequately supplied. If demand exceeds the actual capacity of local governments, leaders will adopt the logic of responsibility avoidance [32], and, accordingly, governance will not necessarily be more effective. The performance evaluation system and bureaucratic structure cause local government officials to pay more attention to outwardly visible performance-related affairs than to governance efficiency. Local government leaders then exhibit excessive enthusiasm towards environmental protection issues, viewing them as means to demonstrate loyalty to higher-level government leaders, which leads to a separation of governance practice from policy objectives and the loss of governance efficiency [33]. Therefore, as various complex governance issues worsen, local government leaders become ineffectual, thereby reducing the efficiency of policy tools.

In the early stages of environmental governance, local government officials’ attention resources are sufficient, and the matching human, material, and financial resources support government attention in completing the corresponding environmental governance goals. Such early investment and strong governance raises expectations, but it also makes it difficult for leaders to judge when and how to adjust their attention allocation. When GEA reaches a critical value, local government officials no longer make problem-oriented decisions but instead make results-oriented decisions. In other words, local governments that are excessively attentive to a certain issue will eventually adopt responsibility-avoidance behaviors. This leads to a serious deviation from policy objectives, which in turn results in negative incentives for environmental governance efficacy.

3. Research Methods and Variables

3.1. Methodology

Spatial econometric methods have been increasingly used in environmental research. In this study, the annual average PM_2.5 level of a city was used to measure the effectiveness of environmental governance. The existing studies confirm that air pollution often exhibits strong spatial spillover due to temperature inversion, wind direction, and other reasons. In other words, the air pollution level of a city may be affected by surrounding cities to some extent [34]. This section discusses spatial matrix selection, spatial autocorrelation testing, and model construction.

3.1.1. Spatial Weight Matrix

The spatial weight matrix is a crucial component in spatial modeling, as it reflects the interdependence among individuals in a given space and describes the level of mutual influence among spatially situated individuals. In this study, we referred to the existing theoretical literature to construct two types of spatial weight matrices: the 0–1 adjacency matrix (W1), which is used in benchmark regression; and the inverse distance spatial weight matrix (W2).

The W1 was based on the assumption that a city only has an interactive relationship with its adjacent areas, and mainly considers the mutual influence of spatial adjacent units. The specific representation was as follows: all the diagonal elements of the spatial weight matrix are 0, and the other elements are $Wij$, indicating the spillover intensity between a pair of spatial units. If city i borders city j, $Wij$ = 1, otherwise, it is 0.

$$Wij=\{\begin{array}{l}1,\hspace{0.33em}\hspace{0.33em}region\hspace{0.33em}i\hspace{0.33em}is\hspace{0.33em}adjcent\hspace{0.33em}to\hspace{0.33em}region\hspace{0.33em}j\\ 0,\hspace{0.33em}\hspace{0.33em}region\hspace{0.33em}i\hspace{0.33em}is\hspace{0.33em}not\hspace{0.33em}adjcent\hspace{0.33em}to\hspace{0.33em}region\hspace{0.33em}j\end{array}\hspace{0.33em}(i\ne j)$$

Geospatial spatial weight matrices are based on the effects of interactions between different spatial units. According to the first law of geography, the closer the distance, the greater the weight influence. We used the inverse distance matrix, denoted as the W2.

$$Wij=\{\begin{array}{l}1/d_{ij},i\ne j\\ 0,\hspace{0.33em}\hspace{0.33em}i=j\end{array}$$

3.1.2. Spatial Autocorrelation Test

Moran’s I was calculated with the W1 and the annual mean value of PM_2.5 in each city, and its value was usually in the interval [−1, 1]. If Moran’s I was greater than 0, it indicated that there was a positive spatial correlation between air pollution in different regions; otherwise, it indicated that there was a negative spatial correlation. As shown in

Table 1. Results of global Moran’s I statistics.

Year	Moran’s I	z	p-Value
2011	0.761 ***	20.515	0.000
2012	0.762 ***	20.543	0.000
2013	0.785 ***	21.186	0.000
2014	0.725 ***	19.583	0.000
2015	0.819 ***	22.095	0.000
2016	0.811 ***	21.881	0.000

Note: *** p < 0.01.

, the results for the six-year period are statistically significant at the 1% level, and all show a positive correlation, suggesting a positive spatial relationship between air pollution levels in each region.

To further investigate the spatial distribution of air pollution levels in different regions, we calculated the local Moran’s I index. As depicted in Figure 1, a comparison of the scatter plots from 2012 and 2015 revealed that the majority of the Moran scatter points are clustered in quadrants I and III, indicating a positive correlation between air pollution levels in neighboring areas. Specifically, at the geographical level, the Yangtze River Delta Economic Zone and the Beijing–Tianjin–Hebei region are primarily clustered in the first quadrant (high–high region), while the western region is mainly clustered in the third quadrant (low–low region). Comparing the data over time, the clustering patterns in 2015 are more pronounced than those in 2012, providing further evidence of the spatial correlation in air pollution.

To illustrate the spatial agglomeration relationship and trend in air pollution over time, we created a visual representation of the PM_2.5 concentration in 285 cities from 2011 to 2016. As depicted in Figure 2, there was a gradual decrease in air pollution levels over the years, indicating the effectiveness of environmental governance efforts. Geographically, the most heavily polluted areas are concentrated in the Beijing–Tianjin–Hebei region and the Yangtze River Delta urban area, which aligns with the scatter distribution pattern observed in the local Moran scatter plot.

3.1.3. Spatial Econometric Model

We conducted a series of statistical tests on the regression results of GEA and air pollution (see

Table A1. Evaluation of spatial measurement model selection.

Test Parameters	GEA
LM(lag)test	603.345 ***
Robust LM(lag)test	49.627 ***
LM(error)test	1571.537 ***
Robust LM(error)test	1017.818 ***
Wald_spatial_lag	26.73 ***
LR_spatial_lag	27.81 ***
Wald_spatial_error	31.96 ***
LR_spatial_error	317.90 ***
Hausman test	393.37 ***
LR test	Fixed effect	Statistics
	Individual fixation	23.75 ***
	Time fixation	4575.50 ***

Note: *** p < 0.01.

). The results confirm that the SDM is a better choice. The specific model is shown in Formula (1),

$$P{M}_{2.5it}={\alpha }_0+\delta WP{M}_{2.5it}+{\alpha }_{1}GE{A}_{it}+{\alpha }_2{GEA\_sq}_{it}+{\alpha }_{3}C{V}_{it}+{\alpha }_{4}WGE{A}_{it}+{\alpha }_{5}WGEA{\_sq}_{it}+{\alpha }_{6}WC{V}_{it}{+u}_i+v_t+{\epsilon }_{it}$$

(1)

in which $i$ and $t$ respectively represent city and year; ${PM}_{2.5it}$ is our dependent variable, which refers to the annual average of PM_2.5 in each city; ${WPM}_{2.5it}$ is the interaction term for the dependent variable and the dependent variable of adjacent units; and $GE{A}_{it}$ is the core independent variable, and ${GEA\_sq}_{it}$ is its square, which is used to verify the nonlinear impact of the GEA on air pollution. $C{V}_{it}$ represents the control variables; $W$ represents the W1; $u_i$ and $v_t$ represent individual fixed effects and time fixed effects respectively; and ${\epsilon }_{it}$ is the random disturbance term.

3.2. Data and Variables

Our data comprised 285 prefecture-level cities and above, including four municipalities directly under the central government, for a period ranging from 2011 to 2016. It included data on air quality, the GEA, and socioeconomic characteristics variables at the city level. The air quality data were from Columbia University’s Socioeconomic Data and Applications Center “https://sedac.ciesin.columbia.edu/ (accessed on 14 August 2024)”. GWRs were used to measure GEA. Other variables in the empirical analysis were collected from the China City Statistical Yearbook, the China Industry Economy Statistical Yearbook, and the China Energy Statistical Yearbook. The following section introduces the definitions and measurement methods of the primary variables used in this paper.

3.2.1. Dependent Variable: Environmental Governance Efficiency

We used the annual average concentration of PM_2.5 as a proxy variable for environmental governance efficiency. Based on the global surface PM_2.5 concentration data provided by Columbia University, we identified the annual average PM_2.5 concentration in prefecture-level cities in China over time and constructed a set of panel data. Specifically, we obtained 1710 yearly, city-level samples.

3.2.2. Independent Variables: GEA

Existing research often uses general sentiment dictionaries to quantify government officials’ attitudes [35,36]. However, these methods may not accurately capture the linguistic features of policy texts. Therefore, this study constructed and used a degree term dictionary that was tailored for policy texts using machine learning methods and applied degree terms to weight the environmental protection keywords and measure GEA. The construction process is detailed in Figure 3 and was divided into three steps: seed word selection, degree word expansion, and the measurement of GEA. After text preprocessing, we selected sample texts from 1% of the 1710 GWRs to read and identify degree words to assign strength values. After consolidating the degree words from the sample, repeated words were removed to obtain a seed word set. To ensure the effectiveness of the dictionary, it was important to increase the number of words. To achieve this, we used the Continuous Bag-of-Words (CBOW) model in Word2vec [37], training the model on 4470 GWRs that were manually collected by our team to expand our seed words and construct a large corpus of degree words. The specific model is represented by Formula (2),

$$\max \sum w\in D \log p(w|Context(w))$$

(2)

where $D$ is the training set corpus, $w$ represents the central word, and $Context(w)$ represents the context of the central word. This model was trained on a large number of government corpora, so similar words that were suggested were more relevant in the context of government policy texts. We screened out words with a similarity greater than 0.8 and a word frequency greater than 500. This removed words that were not closely related to our seed words or that were not used frequently by the governments. By repeating this step until no new words appeared, we finally generated a degree word dictionary containing 494 words (see

Table A3. Examples of degree words and their score values.

Degree Words in English	Degree Words in Chinese	Part of Speech	Score Values
Quicken	加快	Verb	1
Actively	积极	Adverb	1
Lead	引领	Verb	1
Establish	设立	Verb	1
Deepen	深化	Verb	2
Emphatically	重点	Adverb	2
Increase	加大	Verb	2
Reinforce	加强	Verb	2
Comprehensively	全面	Adverb	3
Significant	重大	Adjective	3
Unprecedented	空前	Adjective	3
Thoroughly	彻底	Adverb	3

). In this way, we effectively avoided the weak correlations that can result from using general word-like tools, such as custom dictionaries and synonyms.

This study utilized environmental keywords that were identified in previous research [9,38], as shown in

Table A4. Keywords selected in this study.

Keywords in English	Keywords in Chinese
PM10	PM10
PM2.5	PM2.5
ozone	臭氧
atmospheric pollution	大气污染
nitric oxide	氮氧化物
low-carbon development	低碳发展
sulfur dioxide	二氧化硫
environmental protection	环保
environment	环境
environmental pollution	环境污染
environmental quality	环境质量
environmental governance	环境治理
environmental impact assessment	环评
motor vehicle exhaust	机动车尾气
centralized heating	集中供热
emission reductions	减排
sustainable development	可持续发展
renewable energy sources	可再生能源
air quality	空气质量
blue sky	蓝天
afforestation	绿化
green development	绿色发展
energy consumption	能耗
clean energy	清洁能源
ecological environment	生态环境
fine particulate matter	细颗粒物
new energy	新能源
raise dust	扬尘
air pollution control	治气
pollution control	治污

. In order to more accurately identify the government’s semantic intensity, we set the degree word recognition range of the environmental keywords to 5-word intervals on either side of the keywords to determine the semantic strength of the government. The specific calculation process is shown in Formula (3), where $GE{A}_{it}$ is our independent variable (GEA) for city $i$ in year $t$; ${\mathit{Keyword}}_{ipt}$ is the frequency of the keyword, where $p$ stands for different keywords; and ${\mathit{strength}}_{ipt}$ is the semantic strength of the keyword.

$$GE{A}_{it}={\displaystyle \sum _{p=1}^n}{\mathit{Keyword}}_{ipt} \times {\mathit{strength}}_{ipt}$$

(3)

3.2.3. Control Variables

The control variables in this study mainly included the economic development level (Ln(PGDP)), population size (Ln(POP)), fiscal environmental expenditure (Ln(Fis expenditure)), urban registered unemployment rate (Unemploy), fiscal revenue (Ln(FR)), industrial structure (Second), technology capital input (TCI), and human capital input (HCI). Considering that motor vehicle exhausts and coal burning are important causes of air pollution, the control variables also included coal consumption (Ln(Coal)) and gasoline consumption (Ln(Gas)). However, there were some provincial indicators from the China Environment Yearbook, such as coal consumption, gasoline consumption, and fiscal environmental expenditure, and they needed to be further processed into city-level indicators. Air pollution is directly related to the scale of regional industry; the higher the proportion of regional industrial output, the more industrial pollution will be generated, leading to higher expenditures on environmental control in times of high pollution [39]. Therefore, drawing on the existing literature, we adopted an approach similar to calculating the level of product complexity [40], using the proportion of industrial output at the prefecture level versus the total industrial output at the provincial level as weights, and multiplying by prefecture-level variables to convert provincial variables into prefecture-level indicators.

Table 2. Descriptive statistical analysis.

Variables	Definitions	N	Mean	S.D.	Min	Max
PM2.5	Annual PM2.5 concentration (μg/m3)	1710	47.06	15.49	16.13	108.5
GEA	Attention strength index	1710	44.78	15.81	8	123
GEA_sq	The square of attention strength index	1710	2255	1666	64	15,129
Ln(PGDP)	Logarithm of per capita GDP	1710	10.60	0.582	8.773	13.06
Ln(POP)	Logarithm of population	1710	5.885	0.691	2.970	8.129
Ln(Fis expenditure)	Logarithm of fiscal expenditures on environmental protection	1710	11.20	0.920	7.969	15.11
Unemploy	Proportion of unemployment (%)	1710	5.615	3.001	0.297	29.58
Ln(FR)	Logarithm of local fiscal revenue	1710	13.88	1.041	11.29	17.98
Second	Proportion of secondary industries to regional GDP	1710	49.15	10.31	14.95	89.34
TCI	Proportion of technology expenditures in total fiscal expenditure (%)	1710	1.548	1.518	0.0677	20.68
HCI	Proportion of human capital investment in total fiscal expenditure (%)	1710	18.29	4.102	1.038	35.62
Ln(Coal)	Logarithm of coal consumption	1710	16.01	1.018	12.39	18.31
Ln(Gas)	Logarithm of gasoline consumption	1710	12.43	1.008	7.442	15.67

presents the descriptive statistics for the variables used in this study. As shown above, the statistical results of the air pollution indicators revealed significant differences between cities. For instance, the average PM_2.5 level (47.06 μg/m³) exceeded the WHO standard of 35 μg/m³, with a minimum value of 16.13 μg/m³, a maximum value of 108.5 μg/m³, and a standard deviation of 15.49 μg/m³. Moreover, the variance inflation factors between the core independent variable and the other independent variables were less than 10, so there is no multicollinearity problem (see

Table A2. Variance inflation factor.

Variables	VIF	1/VIF
GEA	1.16	0.860
Ln(PGDP)	7.23	0.138
Ln(POP)	6.1	0.164
Ln(Fis expenditure)	6.39	0.156
Unemploy	1.23	0.814
Ln(RE)	9.32	0.107
Second	1.58	0.63
TCI	1.66	0.602
HCI	1.47	0.681
Ln(Coal)	4.1	0.244
Ln(Gas)	7.09	0.141
MeanVIF	4.3	-

).

4. Empirical Results

4.1. Background on the Structure and Functioning of Local Governments in China

In China’s administrative hierarchy, local governments exist at multiple levels, including provinces, cities, and counties, and each of these governments has specific responsibilities and authorities [41]. Currently, China’s institutional design combines centralization with local economic decentralization. Specifically, while the central government delegates certain administrative powers, it retains significant control over local governments through mechanisms such as fiscal allocations, personnel appointments, and policy directives [30]. The “11th Five-Year Plan for National Environmental Protection”, released in 2006, was the first to explicitly include local government environmental protection work in the performance appraisal system in order to better coordinate economic development with environmental issues across regions. By delegating economic and environmental governance powers to local governments, the central government intended to create diverse policy responses and encourage local initiatives in environmental governance [42]. Local governments’ interpretations of environmental campaigns are sometimes influenced by officials’ personal interests, which can ultimately lead to deviations in their implementation and the governance outcomes [26].

4.2. Baseline Regression

Table 3. Regression results of the spatial econometric model.

Variables	(1) Two-Way Fixed Effects Model	(2) SAR	(3) SEM	(4) SDM
	PM2.5	PM2.5	PM2.5	PM2.5
GEA	−0.0791 ***	−0.0347 ***	−0.0207 **	−0.0312 ***
	(0.029)	(0.012)	(0.010)	(0.012)
GEA_sq	0.0007 ***	0.0004 ***	0.0002 **	0.0003 ***
	(0.000)	(0.000)	(0.000)	(0.000)
W × PM2.5	-	0.9170 ***	1.1777 ***	0.9137 ***
	-	(0.009)	(0.002)	(0.009)
W_x	-	-	-	YES
Control variables	YES	YES	YES	YES
Spatial fixed effects	YES	YES	YES	YES
Temporal fixed effects	YES	YES	YES	YES
Observations	1710	1710	1710	1710
R-squared	0.489	0.320	0.266	0.338
Sigma2	-	2.1933 ***	2.0282 ***	2.1554 ***
	-	(0.080)	(0.069)	(0.078)
LogL.	-	−3350.4557	−3495.4968	−3336.5487

Note: Standard errors in parentheses. *** p < 0.01, ** p < 0.05. Due to space limitation, the results of the spatially lagged independent variables and control variables are not reported here and are readily available upon request.

presents the results of the benchmark regression, with all models including geographic and time fixed effects. Initially, a two-way fixed effects model was used to estimate the relationship between GEA and environmental governance efficiency. As shown in Column (1), without considering spatial spillover effects, there was a U-shaped relationship between GEA and air pollution. This finding aligns with previous perspectives, suggesting that, in the context of environmental campaigns, local governments can effectively improve air quality in the short term by increasing their focus on environmental protection [43,44]. However, in the long term, excessive attention to environmental issues by local governments can lead to a decline in governance efficiency. The U-test, a critical method for verifying nonlinear relationships [45], revealed that the regression curve reached an inflection point at a GEA value of 54.46, which falls within the range of values (8 to 123). The slope of the regression line to the left of this inflection point was −0.0675, while the slope to the right was 0.0996, providing further evidence of the nonlinear relationship.

After considering the impact of spatial spillover, we used the Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), and SDM for the regression analysis. Regardless of which model we used, the coefficients of the spatial lag variable W × PM_2.5 in Columns (2), (3), and (4) were found to be significantly positive at the 1% level, indicating the existence of a spatial spillover effect associated with air pollution. According to Column (4), the impact coefficient of GEA on air pollution was −0.0312, and the impact coefficient of the square term of GEA on air pollution was 0.0003, which are significant at the 1% level. This indicates that the impact of GEA on the efficiency of environmental governance is nonlinear, and that excessive GEA will reduce the efficiency of environmental governance.

In the early stages of environmental governance, local governments tend to have high enthusiasm and invest heavily in environmental protection, leading to strong policy implementation. During this period, the allocation of resources towards environmental protection is reasonable, resulting in a reduction in emissions [46,47]. As a result, the increased GEA reduces air pollution. However, when GEA becomes too high, and exceeds their actual capacities, officials may stop allocating limited time and resources to policy implementation. Instead, they may prioritize completing convoluted paperwork that circumvents the objectives of policy execution, solely pursuing formal compliance [30]. This can ultimately result in a decrease in the effectiveness of environmental governance.

Considering the marginal impact of changes in the independent variables, we further calculated the direct and indirect impacts of the regression. The results are shown in

Table 4. Average direct and indirect effect estimates.

Variables	Direct Effects	Indirect Effects
	(1) PM2.5	(2) PM2.5
GEA	−0.0571 **	−0.4543 *
	(0.023)	(0.273)
GEA_sq	0.0006 ***	0.0041 *
	(0.000)	(0.003)
Ln(PGDP)	0.3651	−12.9861 **
	(0.621)	(6.304)
Ln(POP)	−1.5538	−22.8878
	(2.522)	(33.223)
Ln(Fis expenditure)	−0.1322	3.7411
	(0.538)	(4.115)
Unemploy	0.0065	0.2539
	(0.046)	(0.515)
Ln(RE)	−1.1792 **	−9.2447 *
	(0.523)	(5.469)
Second	0.0180	0.6751 **
	(0.029)	(0.313)
TCI	−0.0957	−1.1866
	(0.088)	(1.058)
HCI	0.0472	0.9806 **
	(0.039)	(0.449)
Ln(Coal)	2.5162 ***	11.2579 **
	(0.615)	(5.448)
Ln(Gas)	−1.0908 **	−2.0337
	(0.534)	(3.462)

Note: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. The expected direct effect of GEA on air pollution showed a U-shaped relationship at the 5% significance level, while the expected indirect effect showed a U-shaped relationship at the 10% significance level.

4.3. Robustness Check

We replaced the 0–1 adjacency matrix, W1, with the inverse distance spatial matrix, W2. The results of the regression analysis are presented in Column (1) of

Table 5. Alternative dependent and independent variable.

Variables	(1)	(2)	(3)	(4)	(5)
	PM2.5	PM2.5	PM2.5	PM2.5_was	PM2.5_max
GEA	−0.0502 ***	-	-	−0.0192 *	−0.0407 **
	(0.018)	-	-	(0.011)	(0.018)
GEA_sq	0.0005 ***	-	-	0.0002 **	0.0005 ***
	(0.000)	-	-	(0.000)	(0.000)
GEA_dic	-	−1.2784 ***	-	-	-
	-	(0.462)	-	-	-
GEA_dic_sq	-	0.3866 ***	-	-	-
	-	(0.119)	-	-	-
GEA_sentence	-	-	−0.0913 *	-	-
	-	-	(0.049)	-	-
GEA_sentence_sq	-	-	0.0079 **	-	-
	-	-	(0.003)	-	-
W × PM2.5	2.7645 ***	0.9143 ***	0.9155 ***	0.9136 ***	0.8535 ***
	(0.029)	(0.009)	(0.001)	(0.009)	(0.012)
W_x	YES	YES	YES	YES	YES
Control variables	YES	YES	YES	YES	YES
Spatial fixed effects	YES	YES	YES	YES	YES
Temporal fixed effects	YES	YES	YES	YES	YES
Observations	1710	1710	1710	1710	1710
R-squared	0.061	0.330	0.334	0.417	0.352
sigma2	4.7109 ***	2.1248 ***	2.1584 ***	1.8296 ***	5.0132 ***
	(0.162)	(0.076)	(0.181)	(0.067)	(0.181)
LogL.	−3779.888	−3335.400	−3339.345	−3191.750	−4002.658

Note: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

, indicating a significantly positive spatial lag coefficient. The estimated coefficient for the relationship between GEA and air pollution was consistent with previous results and was significant at the 1% level. We replaced the original core independent variable with the GEA that was measured using the dictionary method (see Column (2)). In addition, we utilized sentence frequency to measure GEA (see Column (3)). We also replaced the dependent variable. We used PM_2.5 data that were published by the Atmospheric Composition Analysis Group of Washington University in St. Louis “https://sites.wustl.edu/acag/datasets/surface-pm2-5/ (accessed on 14 August 2024)” to measure air pollution (see Column (4)). Existing research also used the extreme value of annual pollution data as a way to measure pollution levels [48]. Results are shown in Column (5). The relationship between GEA and the efficiency of environmental governance remained robust.

Table 6. Additional robustness test.

Variables	(1)	(2)	(3)
	PM2.5	PM2.5	PM2.5
GEA	−0.0295 **	−0.0293 **	−0.0299 **
	(0.012)	(0.012)	(0.012)
GEA_sq	0.0003 ***	0.0003 ***	0.0003 ***
	(0.000)	(0.000)	(0.000)
Ln(SO2)	0.3335 ***	-	-
	(0.115)	-	-
Ln(water)	−0.1751	-	-
	(0.128)	-	-
Ln(smoke)	−0.0476	-	-
	(0.087)	-	-
Ln(use_rate)	0.0074 **	-	-
	(0.003)	-	-
pilot	-	-	0.4582 **
	-	-	(0.210)
W × PM2.5	0.9115 ***	0.9079 ***	0.9097 ***
	(0.009)	(0.009)	(0.009)
W_x	YES	YES	YES
Control variables	YES	YES	YES
Spatial fixed effects	YES	YES	YES
Temporal fixed effects	YES	YES	YES
Observations	1710	1686	1710
R-squared	0.427	0.353	0.422
sigma2	2.1161 ***	2.1794 ***	2.1271 ***
	(0.076)	(0.080)	(0.076)
LogL.	−3321.975	−3297.509	−3330.039

Note: Standard errors in parentheses. *** p < 0.01, ** p < 0.05.

shows the result of additional robustness. Although the benchmark regression included variables that may affect air pollution, we added additional variables related to air pollution to avoid any potential missing factors, including industrial sulfur dioxide and wastewater emissions, industrial soot emissions, and the comprehensive utilization rate of general industrial solid waste (see Column (1)). Due to the hierarchical structure of China’s administrative divisions, which consist of central, provincial, and prefecture-level cities, there are four municipalities (Beijing, Shanghai, Tianjin, and Chongqing) that are directly governed by the central government and are equivalent to provinces in administrative rank. Ordinary prefecture-level cities, on the other hand, are subordinate to provincial governments. To further eliminate potential interference from administrative factors, this study excluded these municipalities, which are more developed than prefecture-level cities (see Column (2)). In addition, the government has also carried out local trials to protect the environment. Specifically, in 2012, China established a group of 74 cities to be the first to implement new air quality standards. These cities have established monitoring sites to conduct air quality testing, publish real-time monitoring data for six basic items (SO₂, NO₂, PM₁₀, O3, CO, and PM_2.5), and track the AQI. The disclosure of air quality information may force local governments to change their actions and priorities. Therefore, to eliminate the possible impact of these events on environmental governance, we further controlled for the pilot cities used in the model (see Column (3)). The above tests show the robustness of our results.

5. Conclusions and Policy Implications

Based on the hypothesis of the bounded rationality of government officials, this paper introduces China’s pressure-driven system into the research of government attention for the first time and provides an explanation for the perspective that increased attention enhances the organization’s ability to identify problems but does not guarantee ideal governance outcomes. The significant contribution of this paper is to improve the measurement method of government attention and provide a new set of measurement procedures. Most previous studies used the dictionary method, but the disadvantage is that they ignored the differences in the attitudes of different local governments towards the same political affairs. We used policy semantic intensity to weight government attention, thereby improving the accuracy of this indicator. Furthermore, using air pollution as an example, we used SDM to verify the U-shaped relationship between changes in GEA and the effectiveness of environmental governance. This provides a theoretical basis for the government to allocate attention rationally to achieve sustainable environmental governance.

The direct approach of campaign-style environmental governance is to incorporate environmental protection into the local government officials’performance evaluation system. This approach incentivizes local officials to pay greater attention to environmental issues, leading them to reduce investments in economic development and reallocate funds to environmental protection to achieve higher performance ratings [49]. However, as various levels of government increase their funding for environmental protection projects, a lack of funding can become problematic, thereby reducing the efficiency of environmental governance [50]. Therefore, it can be inferred that GEA may influence environmental governance efficiency through government funding. However, verifying the potential impact of government funding through quantitative research can be challenging due to unobservable factors, such as the efficiency of fund usage and the behavior of implementers. Future research will require qualitative methods, such as field surveys, to validate these effects.

According to the conclusions, this study has policy implications for improving the efficiency of environmental governance. Firstly, the biggest problem faced by local governments when dealing with environmental issues is how to optimize the structure of attention allocation, rather than insufficient attention. To adapt to the complexity and long-term nature of environmental governance, dynamic assessments based on initial conditions and resource endowment should be implemented in addition to increasing the focus on evaluating environmental protection effects. Secondly, to prevent the spillover effect of environmental pollution and promote regional collaborative environmental governance, regional governance performance should be considered a key indicator in assessing officials’ work effectiveness and abilities. This will enhance officials’ enthusiasm for participating in regional collaborative efforts and improve environmental protection effectiveness. Finally, the central government should establish and strengthen institutional designs for the effective allocation of attention resources, including frameworks for the long-term prevention and supervision of environmental pollution, and guide local government leaders in allocating attention resources through appropriate incentive mechanisms.

Nonetheless, there are still some limitations to this study: First, while this study identified a nonlinear relationship between GEA and environmental governance effectiveness, it did not test the specific mechanisms involved. We have only discussed the potential role of funding allocation, which can be further examined in future research. Second, due to data availability, the time span of this study could not be further expanded. We could only examine the short-term effect of GEA on the efficiency of environmental governance. In addition, this study only provided a theoretical analysis and empirical evidence as it pertains to environmental governance and government attention in China. We did not further verify the existence of a nonlinear relationship between GEA and the efficiency of environmental governance in other countries; future research could assess this relationship with other data.