Evaluating the Sustainability Impact of Ship Emission Control Area Policies on Air Quality in Inland Yangtze River Cities and Underdeveloped Hainan Coastal Cities

Abstract

Ship emissions represent a significant challenge to environmental sustainability and public health. The widely implemented Emission Control Area (ECA) policy aims to mitigate these emissions; however, existing research often overlooks inland and underdeveloped coastal regions. This study evaluates the impact of China’s Domestic Emissions Control Area (DECA) policy on SO₂ and PM_2.5 air concentrations in three inland cities along the Yangtze River and three underdeveloped coastal cities since its inception in 2018. Employing regression discontinuity (RD) and difference-in-differences (DID) methodologies, this analysis seeks to provide a comprehensive understanding of the DECA’s effects on air quality and its implications for sustainability in these areas. This study revealed that DECA policies resulted in significant improvements in both SO₂ and PM_2.5 concentrations, contributing directly to environmental sustainability and potential public health co-benefits in inland urban areas as well as underdeveloped coastal cities. In the case of inland cities, the daily average concentrations of SO₂ in Yichang and Taicang decreased by 5.3% and 18.9%, respectively, while the average daily concentrations of PM_2.5 saw reductions of 21.9% and 13.9%, respectively. Among the underdeveloped coastal cities, the average daily concentration of SO₂ in Haikou, Danzhou, and Sanya declined by 1.6%, 20.2%, and 21.2%, respectively; additionally, the average daily concentrations of PM_2.5 in Danzhou and Sanya decreased by 13.8% and 9.5%, respectively. The effectiveness of this policy exhibited geographical variation within inland cities and was influenced by urban development indicators in coastal areas. These findings not only underscore the success of the DECA policy in enhancing air quality but also highlight its role in advancing sustainable development goals. They provide essential evidence for formulating effective and sustainable emissions control strategies applicable to similar inland regions and underdeveloped coastal settings worldwide.

1. Introduction

1.1. The Rise of the Shipping Industry and Environmental Challenges

The maritime industry, which accounts for over 80% of international cargo transportation, serves as a cornerstone of the global economy [1,2]. However, the expansion of maritime activities has exacerbated air quality in proximity to ports due to ship emissions [3]. Air pollution adversely affects human health in various ways, including respiratory diseases, developmental issues, and increased morbidity rates [4,5,6,7,8]. Addressing this issue is therefore critical for achieving sustainable development, which seeks to balance economic growth with environmental protection and public health. Consequently, mitigating ship emissions has become a key focus for promoting sustainable practices within the maritime sector. In response to these challenges, researchers have investigated the application of machine learning techniques for air quality prediction. This endeavor aims to provide a scientific foundation for evidence-based interventions designed to enhance air quality [9].

Ship emissions primarily consist of sulfur oxides (SO_X), nitrogen oxides (NOx = NO + NO₂), and particulate matter (PM) [10,11,12,13,14]. Notably, SO_X released from ship engine combustion contributes approximately 13% of global anthropogenic SO₂ emissions [15,16]. The SO_X emitted into the atmosphere forms sulfate (SO₄) aerosols that significantly impact human health and contribute to acidification in both terrestrial and aquatic environments [17]. According to research conducted by the Maritime Environment Protection Committee, premature deaths attributable to ship-emitted SO_X are projected to exceed 570,000 between 2020 and 2025 [18]. Furthermore, exposure to PM_2.5 particulate matter poses serious threats to human health [19,20]. The World Health Organization indicates that each increase of 10 μg/m³ in PM_2.5 levels correlates with a rise in mortality risk ranging from 4% to 26% [21].

1.2. International Regulation on Ship Emissions

The sulfur content in fuel has a direct impact on the emission levels of sulfur oxides (SO_X) and an indirect effect on the emissions of PM_2.5 particulates [22,23]. The use of low-sulfur fuel can significantly reduce pollution resulting from ship emissions [24,25,26]. In response to the detrimental effects of ship emissions, the International Maritime Organization (IMO) ratified the International Convention for the Prevention of Pollution from Ships (MARPOL) in 1997, which was subsequently revised in 2008 [27]. This regulation established Emission Control Areas (ECAs), including regions such as the Baltic Sea, North Sea, North American coastal areas (including those of the United States and Canada), and U.S. Caribbean Sea area, thereby limiting sulfur content in marine fuels within these coastal zones. Since 2015, vessels entering ECAs have been mandated to utilize fuel with a sulfur content not exceeding 0.1% (m/m) [28], while recent regulations for other open sea areas require that fuel contain less than 0.5% sulfur (m/m) [18]. Winebrake et al. (2009) concluded that if vessels operating along coastlines transitioned to low-sulfur fuel (<0.5% m/m), it could prevent approximately 33,500 premature deaths globally each year [29].

1.3. Emission Control in China

Despite the significant reduction in air pollution resulting from the implementation of the ECA policy, international shipping emissions control zones do not encompass East Asian waters. Liu et al. (2016) estimated that in 2013, shipping emissions caused 24,000 premature deaths in East Asia, with mainland China being the most affected, accounting for approximately 18,000 deaths (18,000 ± 8600) [30]. In 2016, China introduced the Domestic DECA policy, which initially encompassed the Pearl River Delta (PRD), Yangtze River Delta (YRD), and Bohai Rim (BR) regions. This policy mandated that vessels utilize fuel with a sulfur content of less than 0.5% (m/m) [31]. Since 2019, the scope of regulations has expanded to include all vessels operating within Chinese territorial waters. Furthermore, stricter sulfur content limits of 0.1% (m/m) have been implemented for ocean-going vessels in designated areas of the Yangtze and West Rivers—tributaries of the Pearl River—as well as in the waters surrounding Hainan Island [32]. Chen et al. (2019) projected that the DECA policy enforced in the Pearl River Delta region could lead to a reduction in SO₂ emissions by 76% and PM_2.5 emissions by 57% before 2030, ultimately resulting in a 30% decrease in mortality associated with ship-emitted PM_2.5 [33].

1.4. Current Research Status and Limitations

Existing assessments of ECA policies, including DECA, have primarily concentrated on developed coastal port regions. For instance, Wan et al. (2019) utilized the DID model to assess the effects of the DECA policy in the Yangtze River Delta (YRD), Pearl River Delta (PRD), and Bohai Rim (BR) regions, thereby confirming its positive impact on air quality in developed coastal port areas [34]. Similarly, Zhou et al. (2023) applied an RD model from econometrics to quantify the policy’s influence on environmental improvements at Shanghai Port, analyzing both local and regional perspectives [35]. Nevertheless, current evaluations of the DECA policy predominantly concentrate on developed coastal cities, resulting in a significant dual gap in our understanding.

First, the effectiveness of DECA policies along major inland waterways (e.g., the Yangtze River) remains largely unexamined. Second, there is a notable scarcity of research regarding the policy impacts in underdeveloped coastal cities, which may encounter challenges distinct from those faced by their developed counterparts (e.g., variations in industrial structure and population demographics). Investigating these areas is critical for two primary reasons: inland river cities bear the burden of ship-related emissions that affect large populations, while underdeveloped coastal cities may be particularly susceptible to air pollution. This vulnerability raises questions about the generalizability of findings derived from major developed ports. Consequently, the existing literature fails to provide a comprehensive understanding of the nationwide environmental benefits associated with DECA policies and their effectiveness across diverse socioeconomic and geographical contexts.

In a study conducted by Cui et al. (2022), the spatial distribution of sulfur dioxide (SO₂) health risk inequality coefficients in China from 2013 to 2017 was examined, revealing that Hainan Province emerged as one of the regions with the highest levels of inequality [36]. Notably, in 2016, premature deaths attributable to PM_2.5 in China accounted for approximately 9.98% of total mortality, with the highest rates observed in provinces such as Anhui, Guangdong, Hebei, and Jiangsu. Furthermore, Guo et al. (2014) and Zhang et al. (2015) indicated that direct emissions, atmospheric transport, and the secondary formation of sulfate and nitrate significantly contribute to the exacerbation of PM_2.5 pollution [37,38]. The methodology selection for this study is grounded in the characteristics of policy implementation. In inland cities, DECA policies exhibit distinct geographical boundaries along the Yangtze River (such as core port sections), which creates a clear breakpoint. Consequently, an RD model design is employed to accurately capture the local average treatment effect of the policy. Conversely, in underdeveloped coastal cities, policy boundaries typically align with administrative borders and lack similar local variations; therefore, a DID model approach is utilized. This method effectively identifies the net effect of the policy by comparing changes in pollutant concentrations (SO₂ and PM_2.5) before and after its implementation between the treatment group (cities with DECA) and a screened control group (similar cities without DECA).

Therefore, this study focuses on three inland cities and three underdeveloped coastal cities in China. The objective is to analyze and demonstrate the impact of the DECA policy on the concentrations of SO₂ and PM_2.5 in the air of both inland and underdeveloped coastal cities by employing RD and DID models. This study offers essential evidence for assessing the effectiveness of DECA policies beyond developed coastal urban areas. In doing so, it establishes a more robust foundation for future policy optimization and facilitates the evaluation of their broader environmental benefits across China’s diverse geographical landscapes.

2. Methods and Data

2.1. DECA Policy and Study Regions

The DECA policy was initially implemented in developed coastal port cities and was subsequently extended to include inland and underdeveloped coastal areas (

Table 1. Schedule of DECA emission control policies.

Region	Fuel Sulfur Content	Effective Date	Policy
PRD, YRD, and BR waters—Conditionally at berth	0.5% (m/m)	From 1 January 2016	MOT (2015) [31]
PRD, YRD, and BR waters—Core ports at berth	0.5% (m/m)	From 1 January 2017
PRD, YRD, and BR waters—All ports at berth	0.5% (m/m)	From 1 January 2018
PRD, YRD, and BR waters—All ports in transit and at berth	0.5% (m/m)	From 1 January 2019
Inland waters for ocean-going vessels	0.1% (m/m)	From 1 January 2020	MOT (2018) [32]
Underdeveloped coastal waters (Hainan Province)	0.1% (m/m)	From 1 January 2022
All coastal waters (to be assessed)	0.1% (m/m)	From 1 January 2025

). Prior to 1 January 2020, the DECA framework encompassed only the waters of the Pearl River Delta (PRD), Yangtze River Delta (YRD), and Bohai Rim (BR). Following this date, its jurisdiction expanded to incorporate the inland waters of both the Yangtze and West Rivers. Within this broadened scope, ocean-going vessels are required to utilize fuel with a sulfur content not exceeding 0.1% (m/m). On 1 January 2022, enforcement of the DECA policy commenced within Hainan Province, marking it as the first underdeveloped coastal region subject to these regulations. Vessels operating in this area must also adhere to the stipulation regarding sulfur levels in fuel. After a comprehensive evaluation process, starting in 2025, other coastal cities will determine whether to adopt regulations that limit sulfur content to no more than 0.1% (m/m) [31].

We selected sample cities located in inland regions and underdeveloped coastal control areas. In the inland water region, cities were chosen from the upstream, midstream, and downstream sections of the Yangtze River DECA zone, specifically Yibin, Yichang, and Taicang (Figure 1). The underdeveloped coastal sample cities—Haikou, Danzhou, and Sanya—were selected from Hainan Province, which was among the first regions to implement a DECA policy. As of 2022, Haikou had a GDP of CNY 213.48 billion with a population of approximately 2.94 million; Danzhou’s GDP was CNY 87.89 billion with a population of around 0.982 million; while Sanya reported a GDP of CNY 84.71 billion and a population of about 1.066 million. All these cities are classified as underdeveloped coastal urban areas.

2.2. Model for Evaluating the DECA Policy

A simplistic approach to policy evaluation typically involves a before-and-after comparison of outcomes. However, this method is inadequate in isolating the effects of the policy from other concurrent economic or environmental trends, thereby compromising the rigor of causal identification. To more accurately ascertain the net effect of the DECA policy, this study employs RD and DID designs—robust “quasi-experimental” methods that effectively address concerns related to endogeneity. At the same time, RD and DID models are extensively employed in econometrics to evaluate the effects of policies [39,40,41].

For the analysis of inland cities, the DECA policy was uniformly implemented across all inland port areas in China, which precludes the establishment of a valid control group comprising unregulated ports. Consequently, an RD design is employed. This model is particularly well-suited for scenarios characterized by a sharp change in treatment assignment at a specific cutoff point (e.g., the implementation date), thereby facilitating the creation of comparable groups situated just above and below this threshold. It is considered one of the methodologies closest to randomized experiments for identifying local causal effects with high internal validity [42,43]. The applicability of the RD model in this context relies on the abrupt policy change that took effect on 1 January 2020, which introduces a clear temporal discontinuity. By comparing samples that are narrowly positioned around this cutoff, we can effectively control for unobserved confounders that exhibit smooth variation over time.

In the analysis of underdeveloped coastal cities, Hainan Province represents a region that has been classified as underdeveloped and was among the first to implement the DECA policy. In contrast, other regions that did not adopt the DECA policy were utilized as control groups (Figure 1). This aligns with the fundamental premise of the DID model: there exists an experimental group that is influenced by the policy and a control group that remains unaffected. Consequently, the DID approach is employed. Its core principle involves utilizing the differential changes over time between the treatment group (subject to the policy) and the control group (not subject to it) in order to account for pre-existing fixed differences between groups as well as common temporal trends. This methodology ultimately provides a more accurate estimate of the policy effect [44].

2.2.1. RD Model for Inland Cities

The RD model is specifically designed to identify and evaluate the impact of other variables when a particular variable undergoes a discontinuity. Fundamentally, it utilizes a continuous variable (such as time) that determines the probability of individuals receiving policy intervention (or treatment) on either side of a critical threshold. The model is categorized into two types: sharp RD (SRD) and fuzzy RD (FRD). In SRD, there is an abrupt transition from zero to one in the treatment probability at an exact breakpoint (x = c). Conversely, FRD exhibits a shift from ‘$a$’ to ‘$b$’ at the breakpoint (x = $c$), where 0 < $a$ < $b$ < 1. In this study, the DECA had a defined implementation date; thus, the likelihood of DECA implementation was zero prior to and one following the cutoff point. Consequently, we adopted the SRD model for our analysis. The RD model utilized in this research [45,46] can be described as follows:

$$Y_{it}^{a,b}={\alpha }_0+\beta f\left(t\right)+r{D}_t+\delta Z_{it}^{a,b}+g\left(\cdot \right)+\epsilon$$

(1)

where $Y_{it}^{a,b}$ is the dependent variable representing the daily average concentration of gas observed in city $i$ at time $t$, and the superscripts $a$ and $b$ represent SO₂ and PM_2.5, respectively; $t$ is the time series, with $t$ being negative if the current time is before the cutoff point and positive otherwise; ${\alpha }_0$ represents the fixed effect; $f\left(t\right)$ denotes a function representing the time trend using a polynomial, with $\beta$ as its estimated coefficient; $D_t$ is the treatment variable (when $t\ge 0$, it takes the value 1, indicating that the policy has been implemented, and 0 otherwise, indicating that the policy has not been implemented); $r$ (coefficient of $D_t$) represents the policy effect, where a negative value suggests that the DECA policy helps reduce the concentration of pollutants; $Z_{it}^{a,b}$ is the set of relevant control variables, with $\delta$ as their coefficients; $g\left(\cdot \right)$ is the interaction term between $f\left(t\right)$ and $D_t$, representing potential time effects during policy changes; and $\epsilon$ is the error term used to explain unknown influences.

The selection of the polynomial order for the time trend function is crucial for obtaining unbiased estimates of the treatment effect. In this study, we determined the optimal polynomial order (e.g., linear, quadratic, or cubic) based on the Akaike Information Criterion (AIC) [47] and the Bayesian Information Criterion (BIC) [48]. These criteria balance model fit and complexity to mitigate overfitting. We estimated models with progressively higher orders and selected the specification that minimized these criteria. Furthermore, to ensure the robustness of our results, we adhered to common practices in econometrics [45,49] by presenting estimates across multiple polynomial orders and bandwidths. This approach demonstrates that our core findings regarding policy effects are not sensitive to these specification choices.

2.2.2. DID Model for Coastal Underdeveloped Cities

The DID model is extensively utilized for assessing the causal impact of policy interventions, owing to its capacity to reduce interference from other policies [50,51]. This model entails dividing panel data into intervention and non-intervention groups, allowing for the inference of policy intervention effects by comparing differences observed before and after the implementation of the intervention [52]. By controlling for pre-existing individual characteristics and time-fixed trends, the DID model interprets any remaining differences as changes attributable to policy interventions. The representation of this model is as follows:

$$Y_{it}^{a,b}={\alpha }_0+{\alpha }_{1}du+{\alpha }_{2}dt+{\alpha }_{3}du\times dt+\delta Z_{it}^{a,b}+{\epsilon }_{it}$$

(2)

To address heteroscedasticity, logarithmic transformations were applied to each explanatory and control variable using the original data. The model employed is outlined as follows:

$$ln{Y}_{it}^{a,b}={\alpha }_0+{\alpha }_{1}du+{\alpha }_{2}dt+{\alpha }_{3}du\times dt+\delta ln{Z}_{it}^{a,b}+{\epsilon }_{it}$$

(3)

where $Y_{it}^{a,b}$ represents the average gas concentration observed in sample city $i$ at time $t$ for the dependent variable, with superscripts $a$ and $b$ representing SO₂ and PM_2.5, respectively. The subscript $t$ denotes the time series, where $t$ is negative before the policy change point and positive afterward; $du$ is a group dummy variable, taking the value of 1 if individual $i$ is affected by the policy implementation (belongs to the treatment group), and 0 otherwise; $dt$ is a policy implementation dummy variable, with a value of 0 before the policy change and 1 after. The interaction term $du\ast dt$ reflects the net effect after the policy change, with a negative coefficient indicating a positive impact of the policy on improving air quality; $Z_{it}^{a,b}$ represents a series of control variables ($\delta$ is the corresponding coefficient); and ${\epsilon }_{it}$ is the random error term.

The other DID model variables are explained below (

Table 2. DID model variables.

	Before Policy Implementation (dt = 0)	After Policy Implementation (dt = 1)	Difference
Treatment group ($\mathit{d}\mathit{u}$ = 1)	${\alpha }_0+{\alpha }_1$	${\alpha }_0+{\alpha }_1+{\alpha }_2+{\alpha }_3$	${\alpha }_2+{\alpha }_3$
Control group ($\mathit{d}\mathit{u}$ = 0)	${\alpha }_0$	${\alpha }_0+{\alpha }_2$	${\alpha }_2$
DID estimation	${\alpha }_1$	${\alpha }_1+{\alpha }_3$	${\alpha }_3$

). This table shows the theoretical framework of the DID model, where $a_0$, $a_1$, $a_2$, and $a_3$ are the parameters to be estimated. The core parameter of concern is $a_3$, which identifies the net effect of the policy by comparing the change of differences between the treatment and control groups before and after the policy, a process that excludes the impact of the initial difference between the groups $a_1$ and the common temporal trend $a_2$.

2.3. Variables and Descriptive Statistics

Data Source: The explanatory variables utilized in this study consist of the daily average concentrations of SO₂ and PM_2.5 in urban air (unit: μg/m³), which were obtained from the China Air Quality Online Monitoring and Analysis Platform (https://www.aqistudy.cn/, accessed on 10 December 2024). In accordance with the varying timelines of policy implementation, the data for the analysis of inland areas covers two distinct periods: from 1 January 2019 to 31 December 2019 (pre-policy implementation) and from 1 January 2020 to 31 December 2020 (post-policy implementation). For the examination of underdeveloped coastal regions, the data spans from 1 January 2021 to 31 December 2021 (pre-policy implementation) and from 1 January 2022 to 31 December 2022 (post-policy implementation). It is essential to acknowledge that the variables utilized in this study exhibit differences in data frequency. This discrepancy in frequency implies that our model may not entirely capture the potential effects of daily fluctuations in macroeconomic variables on air quality. We have adopted a cautious approach regarding this issue in our subsequent analysis and further investigate its possible implications, as well as future enhancements, in the discussion section of Section 5.

Table 3. Descriptive statistics of key variables for Yibin, Yichang, and Taicang cities.

	Variable	Mean	Min	Max	Std.Dev.
I	SO2	8.564	2.000	39.000	4.638
	PM2.5	42.803	5.000	196.000	30.373
	GDP	226.638	194.758	273.557	26.667
	GBE	42.343	26.002	69.085	9.915
	TFV	530.891	458.77	618.152	49.367
	GIO	243.500	186.870	309.807	34.029
	Obs.	731
II	SO2	6.952	3.000	21.000	2.052
	PM2.5	46.450	3.000	284.000	38.358
	GDP	363.590	308.138	457.113	54.009
	GBE	49.440	49.133	49.749	0.309
	TFV	1314.471	1100.788	1571.457	153.911
	GIO	303.399	231.645	387.814	42.958
	Obs.	731
III	SO2	10.245	0.000	32.000	4.936
	PM2.5	28.230	0.000	157.000	22.572
	GDP	113.023	94.676	126.414	10.155
	GBE	13.912	9.420	21.625	3.445
	TFV	167.951	60.780	271.367	43.852
	GIO	201.592	125.800	238.093	34.862
	Obs.	731

Note: I: Yibin City; II: Yichang City; III: Taicang City.

and

Table 4. Descriptive statistics of key variables for Haikou, Danzhou, and Sanya cities.

	Variable	Full				Treatment				Control
		Mean	Min	Max	Std.Dev.	Mean	Min	Max	Std.Dev.	Mean	Min	Max	Std.Dev.
IV	lnSO2	1.823	0.693	3.367	0.619	1.354	0.693	2.485	0.295	2.292	1.099	3.367	0.487
	lnPM2.5	2.651	0.000	4.771	0.665	2.530	0.000	4.771	0.602	2.772	0.693	4.554	0.701
	lnGDP	6.001	5.611	6.496	0.284	6.248	6.059	6.496	0.172	5.754	5.611	5.889	0.099
	lnGBE	3.435	2.139	4.104	0.407	3.165	2.139	3.727	0.378	3.704	3.306	4.102	0.211
	lnTFV	6.947	6.589	7.191	0.115	6.927	6.589	7.191	0.144	6.967	6.820	7.087	0.072
	lnGIO	4.648	3.645	5.762	0.630	4.035	3.645	4.263	0.135	5.260	5.011	5.762	0.160
	Obs.	1460				730				730
V	lnSO2	1.826	0.693	3.091	0.425	1.744	0.693	2.944	0.455	1.909	1.099	3.091	0.376
	lnPM2.5	2.663	0.000	4.718	0.723	2.487	0.693	4.060	0.601	2.839	0.000	4.718	0.790
	lnGDP	4.383	3.354	5.099	0.586	3.888	3.354	4.483	0.422	4.877	4.689	5.099	0.140
	lnGBE	2.573	2.376	2.655	0.115	2.493	2.376	2.609	0.117	2.654	2.652	2.655	0.001
	Obs.	1460				730				730
VI	lnSO2	1.646	0.693	2.708	0.566	1.131	0.693	2.197	0.276	2.162	1.791	2.708	0.182
	lnPM2.5	2.543	0.693	4.277	0.601	2.300	0.693	4.127	0.536	2.787	0.693	4.277	0.563
	lnGDP	4.872	3.985	5.647	0.652	4.235	3.985	4.489	0.175	5.509	5.344	5.647	0.095
	lnGBE	3.127	1.790	3.947	0.502	2.776	1.790	3.774	0.462	3.479	3.165	3.947	0.206
	lnTFV	5.091	2.890	7.019	1.407	3.710	2.890	4.007	0.265	6.473	5.616	7.019	0.256
	lnGIO	3.787	1.359	6.133	1.833	1.977	1.359	2.253	0.259	5.597	4.503	6.133	0.312
	Obs.	1460				730				730

Note: IV: Haikou City; V: Danzhou City; VI: Sanya City. SO₂ and PM_2.5 represent the corresponding daily average concentrations of gases (unit: μg/m³); GDP (unit: billion CNY); GBE (unit: billion CNY); TFV (unit: ten thousand tons); GIO (unit: billion CNY); Obs. denotes the sample size, and the mean, minimum, maximum, and standard deviation represent the mean, minimum, maximum, and standard deviation of the corresponding gas concentrations, respectively. Owing to data applicability, a few monthly data were derived by proportionally extrapolating from quarterly or annual data.

provide descriptive statistics for the primary variables involved in the experiments assessing policy impacts in both inland and underdeveloped coastal regions. Furthermore, control variables were integrated into the experiments to account for the influence of additional factors on gas concentrations. The data presented in Table 3 indicate significant fluctuations among the values of each variable, with notable differences observed in the economic control variables (GDP, GBE, TFV, GIO). These variations are instrumental in assessing their impact on air quality. Table 4 provides descriptive statistics for key variables to facilitate a comparison across the entire sample, as well as between the treatment and control groups. The table reveals that certain fundamental economic variables (such as lnGDP and lnGBE) differ between these groups; thus, it is crucial to account for these variables within the model specifications. Previous research has underscored the correlation between economic development [53], government budget expenditures—particularly those directed towards technology and environmental initiatives [54]—traffic conditions [55], and industrial development [56] in relation to urban air quality. Consequently, the control variables identified for this study include the Gross Domestic Product (GDP), Government Budget Expenditures (GBE), Total Freight Volume (TFV), and Gross Industrial Output Value (GIO). The data utilized were primarily sourced from the statistical portals of the respective city statistics bureaus.

The specific data sources are as follows: Yibin City (http://tjj.yibin.gov.cn/, accessed on 10 December 2024); Yichang City (http://xxgk.yichang.gov.cn/list.html?depid=877&catid=150, accessed on 10 December 2024); Taicang City (http://www.taicang.gov.cn/taicang/dataopen/sjkf.shtml, accessed on 10 December 2024); Haikou City (http://tjj.haikou.gov.cn/, accessed on 10 December 2024); and Danzhou City (https://www.danzhou.gov.cn/danzhou/sj/, accessed on 10 December 2024).

3. Results

3.1. The Improvement in SO₂ Pollution in Inland Areas

The RD model was employed to evaluate the impact of the DECA policy implemented in inland water areas on the daily average concentration of SO₂ in three selected cities (Yibin, Yichang, and Taicang) during 2020 (Figure 2). The figure illustrates the fitted regression curves for the SO₂ concentration before and after the implementation of the policy, with the left side representing the pre-implementation period (1 January 2019 to 31 December 2019) and the right side indicating the post-implementation period (2 January 2020 to 31 December 2020). The green line denotes the fitted curve for the daily average SO₂ concentration prior to policy implementation, while the blue line represents that following its enactment. The effectiveness of this policy was assessed by examining whether there was a significant alteration in the curve upon crossing the designated cutoff point. The average daily concentrations of SO₂ in Yibin City, prior to and following the implementation of the DECA policy, were 9.904 μg/m³ and 4.962 μg/m³, respectively. In Figure 2a, Yibin City’s fitted curve for the daily average SO₂ concentration reveals a marked upward discontinuity. This observation indicates that it is not the DECA policies that have contributed to the improvement in Yibin City’s environmental conditions. Conversely, the average daily concentrations of SO₂ in Yichang City, prior to and following the implementation of the DECA policy, were 7.142 μg/m³ and 6.762 μg/m³, respectively. Figure 2b presents Yichang City’s fitted curve for the daily average SO₂ concentration, which demonstrates a significant downward jump near the cutoff point; this indicates a notable environmental improvement attributable to DECA policy measures within that region. Lastly, the average daily concentrations of SO₂ in Taicang City, prior to and following the implementation of the DECA policy, were 11.261 μg/m³ and 9.135 μg/m³, respectively. Figure 2c depicts Taicang City’s fitted curve for the daily average SO₂ concentration showing a downward shift at the cutoff point. However, the decrease observed is not considerable, suggesting a certain level of effectiveness of the DECA policy in enhancing environmental conditions in that area.

Table 5. The estimation results of SO₂ in inland areas.

Variable	I-SO2		II-SO2		III-SO2
	(1)	(2)	(1)	(2)	(1)	(2)
r	1.397	−0.0242	−2.277 **	−2.554 *	−3.892 **	−3.296 **
	(1.0373)	(0.0656)	(0.9961)	(1.4100)	(1.8693)	(4.3210)
GDP		0.0363		−0.00904		0.113
		(0.0519)		(0.0135)		(0.0222)
GBE		0.0580		−0.00311		−0.0330
		(0.0541)		(0.0152)		(0.0238)
TFV		−0.0274		−0.00753		0.0451
		(0.0169)		(0.0047)		(0.0122)
GIO		−0.00198		0.00607		0.0697 ***
		(0.0332)		(0.0075)		(0.0221)
Controls	No	Yes	No	Yes	No	Yes
Cons.	5.328 ***	−9.426	10.12 ***	14.98 ***	13.32 ***	15.84
	(0.8393)	(11.3302)	(0.5977)	(4.1188)	(1.3424)	(14.7919)
Obs.	730	730	731	731	731	731
R2	0.414	0.448	0.522	0.525	0.090	0.096
AIC	3927.275	3888.181	2590.977	2590.537	4344.490	4343.901
BIC	3968.613	3947.891	2636.921	2654.859	4390.434	4408.223
Order	5	5	6	6	7	7

Note: I-SO₂, II-SO₂, and III-SO₂ represent the analytical results for SO₂ in Yibin, Yichang, and Taicang, respectively. Column (1) displays the results of the model without control variables (the simplified model), whereas column (2) includes all available reference control variables. Values in parentheses indicate standard deviations. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively. The coefficient “r” represents the policy effect coefficient, “cons” represents the constant term, “Obs.” indicates the sample size, “R²” represents the goodness of fit, and “AIC” and “BIC” indicate the Akaike and Bayesian information criteria values, respectively. These criteria are used to determine the order of the polynomial functions.

presents the results of the RD estimation for the daily average concentration of SO₂ in Yibin, Yichang, and Taicang. The fitted curve for SO₂ concentration in Yibin did not indicate a decrease; rather, it showed an increase following the breakpoint. The estimated coefficients with and without control variables were 1.397 and −0.0242, respectively, neither of which exceeded the 90% confidence interval. This suggests that the implementation of the DECA policy did not have a favorable effect on reducing SO₂ concentrations in Yibin. In contrast, for Yichang, the estimated coefficients are −2.277 ** and −2.554 *, which are significant at the 5% and 10% levels, respectively, both before and after including control variables. This indicates that the DECA policy has a positive impact on air quality concerning SO₂ pollution in Yichang City. For Taicang, the coefficients are −3.892 ** and −3.296 **, both significant at the 5% level, suggesting that the effects of ECA policy were more pronounced there compared to other locations. The differing outcomes among these three sample cities may be attributed to geographical variations; specifically, cities closer to coastal inland areas experience more active shipping traffic, which amplifies policy effects. Overall, our findings across these three sample cities indicate a positive improvement in SO₂ levels within inland waterway regions as a result of enforcing DECA policies. Furthermore, analysis from Taicang reveals that the GIO coefficient is significantly positive at the 1% level—indicating that increases in local GIO index values correlate with higher SO₂ concentrations in ambient air.

3.2. The Improvement in PM_2.5 Pollution in Inland Areas

The daily average concentrations of PM_2.5 in Yibin, Yichang, and Taicang before and after the implementation of the DECA policy response were as follows: 45.887 μg/m³ and 39.727 μg/m³ for Yibin, 52.173 μg/m³ and 40.743 μg/m³ for Yichang, and 30.731 μg/m³ and 26.458 μg/m³ for Taicang, respectively. Figure 3a–c illustrates the discontinuity regression curves for PM_2.5 concentrations in Yibin, Yichang, and Taicang, respectively. The graphs indicate a slight decline in the fitted curves of the daily average PM_2.5 concentration near the policy breakpoint across all three sample cities; however, this decline is not particularly pronounced. Given that the ECA policy exerts an indirect influence on local PM_2.5 levels, the graphs may lack a distinct breakpoint. Nevertheless, this absence does not imply that the DECA policy fails to positively affect PM_2.5 improvement; rather, a comprehensive analysis considering overall regression results is warranted.

Table 6. The estimation results of PM_2.5 in inland areas.

Variable	I-PM2.5		II-PM2.5		III-PM2.5
	(1)	(2)	(1)	(2)	(1)	(2)
r	1.397	−20.09	−3.723 ***	−52.01 **	−3.253 **	−27.21 *
	(1.0373)	(21.2906)	(0.8254)	(21.3841)	(1.6169)	(14.9380)
GDP		0.0758		−0.162		−0.119
		(0.1515)		(0.1413)		(0.1212)
GBE		0.174		0.0239		−0.468 **
		(0.1852)		(0.1663)		(0.2269)
TFV		−0.0450		0.0159		0.0605
		(0.0703)		(0.0408)		(0.0679)
GIO		−0.203		−0.192		0.0600
		(0.1413)		(0.1232)		(0.1349)
Controls	No	Yes	No	Yes	No	Yes
Cons.	5.328 ***	144.5 **	10.40 ***	204.4 ***	12.74 ***	105.7 **
	(0.8393)	(62.2232)	(0.5855)	(39.8504)	(1.2821)	(45.6090)
Obs.	730	731	731	731	731	731
R2	0.414	0.473	0.503	0.674	0.094	0.306
AIC	3927.275	6608.2753	2619.794	6596.803	4342.737	6376.326
BIC	3968.613	6672.5971	2656.549	6651.936	4384.086	6440.648
Order	5	5	4	4	5	5

Note: I-PM_2.5, II-PM_2.5, and III-PM_2.5 represent the analysis results in Yibin, Yichang, and Taicang for PM_2.5, respectively. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

presents the results of the RD estimation for the average daily concentrations of PM_2.5 in Yibin, Yichang, and Taicang. In Yibin, the estimated coefficients prior to and following the inclusion of control variables were 1.397 and −20.09, respectively. Although a negative correlation emerged after adding control variables, it did not achieve statistical significance within the 90% confidence interval. This lack of statistical significance suggests that the DECA policy does not have a discernible impact on reducing PM_2.5 concentrations in Yibin’s air quality, which aligns with observed improvements in SO₂ levels in that region. In contrast, for Yichang, the estimated coefficient before incorporating control variables was −3.723 ***, indicating significance at the 1% level. After including these control variables, this coefficient changed to −52.01 **, which is significant at the 5% level. These findings imply that the DECA policy has a substantial positive effect on mitigating PM_2.5 pollution in Yichang. For Taicang, prior to adding control variables, the estimated coefficient was −3.253 **, while post-adjustment, it became −27.21 *, both demonstrating significance at the 5% and 10% levels, respectively; this indicates localized effects attributable to the policy intervention. These findings align with the analysis conducted in the three sample cities concerning SO₂ pollution. Yibin is situated upstream of the Yangtze River control area, where maritime traffic is limited. Consequently, the impact of the DECA policy has been constrained. Furthermore, the estimated coefficient for GBE in Taicang is −0.468 **, indicating that increased government spending positively influences PM_2.5 levels. This relationship can be attributed to the allocation of government expenditures towards environmental protection, technological advancements, and energy conservation initiatives. The other control variables did not demonstrate a significant effect on local PM_2.5 concentrations, as supported by their 90% confidence intervals.

3.3. The Improvement in SO₂ Pollution in Underdeveloped Coastal Areas

The DID method was employed to analyze the implementation of the DECA policy in the waters of Hainan, effective 1 January 2022. This analysis assessed its impact on the daily average concentration of sulfur dioxide (SO₂) in three underdeveloped coastal cities: Haikou, Danzhou, and Sanya, as presented in

Table 7. The estimation results of SO₂ in underdeveloped coastal areas.

Variable	IV-SO2		V-SO2		VI-SO2
	(1)	(2)	(1)	(2)	(1)	(2)
did	−0.872 ***	−0.0646 **	−0.235 ***	−0.137 ***	−0.492 ***	−0.234 ***
	(0.0284)	(0.0293)	(0.0250)	(0.0383)	(0.0255)	(0.0173)
lnGDP		0.155 **		0.301 ***		0.381 ***
		(0.0666)		(0.0786)		(0.0408)
lnGBE		−0.0702 *		−1.250 ***		0.0223
		(0.0374)		(0.405)		(0.0170)
lnTFV		0.424 ***		0.0208 ***		0.0934 ***
		(0.0991)		(0.0049)		(0.0309)
lnGIO		0.826 ***		0.0196 *		0.0384 *
		(0.0339)		(0.0366)		(0.0226)
Controls	No	Yes	No	Yes	No	Yes
Cons.	1.884 ***	0.257	1.885 ***	3.757 ***	1.881 ***	−0.843 ***
	(0.0142)	(0.974)	(0.0125)	(0.706)	(0.0128)	(0.140)
Obs.	1460	1460	1460	1460	1460	1460
R2	0.392	0.590	0.057	0.071	0.203	0.846

Note: VI-SO₂, V-SO₂, and IV-SO₂ represent the analysis results of SO₂ in Haikou, Danzhou, and Sanya, respectively. “did” indicates the policy effect coefficient, and “Controls” indicates whether control variables are included. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

.

The concentrations of SO₂ in Haikou, Danzhou, and Sanya were measured at 3.926 μg/m³ and 3.863 μg/m³, 7.074 μg/m³ and 5.644 g/m³, and 3.564 μg/m³ and 2.808 μg/m³ before and after the implementation of the DECA policy, respectively. Before the inclusion of control variables, SO₂ concentrations in the air across all three sample regions exhibited a significant decrease at the 1% level. Following the incorporation of control variables, the significance levels for the improvement results of the DECA policy in Danzhou and Sanya were found to be 1%, with coefficients of −0.137 *** and −0.234 ***, respectively. The policy effect coefficient for Haikou was recorded at −0.0646 **. Furthermore, in Haikou, Danzhou, and Sanya, the GDP coefficients demonstrated significant positive values at the 5%, 1%, and 1% levels, respectively. In Haikou and Danzhou, the GBE coefficients showed significantly negative values at the 10% and 1% levels, respectively. Additionally, within the Haikou, Danzhou, and Sanya regions, the TFV coefficients were significantly positive at the 1% level; meanwhile, the GIO coefficients displayed significant positivity at both the 1% and two instances at the 10% level.

The negative coefficients and statistically significant associations related to the implementation of DECA policy indicate that the reduction in SO₂ concentrations observed in Haikou, Danzhou, and Sanya is correlated with the enforcement of these DECA policies. Conversely, the positive coefficients for GDP indicate a direct relationship between economic activity and elevated levels of SO₂ pollution. The negative coefficients for GBE in Haikou and Danzhou imply that government financial expenditures in these cities may play a role in alleviating SO₂ pollution. In contrast, the positive coefficients for TFV and GIO signify that increased freight traffic and industrial development are linked to heightened local SO₂ concentrations.

3.4. The Improvement in PM_2.5 Pollution in Underdeveloped Coastal Areas

The results of the DID estimation regarding the causal relationship between fluctuations in PM_2.5 concentrations and the implementation of the DECA policy in underdeveloped coastal cities are presented in

Table 8. The estimation results of PM_2.5 in underdeveloped coastal areas.

Variable	IV-PM2.5	IV-PM2.5	V-PM2.5	V-PM2.5	VI-PM2.5	VI-PM2.5
	(1)	(2)	(1)	(2)	(1)	(2)
did	−0.114 ***	0.0258	−0.353 ***	−0.151 **	−0.399 ***	−0.148 ***
	(0.0401)	(0.0480)	(0.0427)	(0.0647)	(0.0348)	(0.0424)
lnGDP		−0.0564		0.677 ***		0.324 ***
		(0.109)		(0.133)		(0.100)
lnGBE		−0.0647		−2.445 ***		0.189 ***
		(0.0613)		(0.685)		(0.0417)
lnTFV		0.756 ***		0.0028 *		−0.0288
		(0.162)		(0.0144)		(0.0758)
lnGIO		0.227 ***		0.0572 *		−0.0182
		(0.0555)		(0.0572)		(0.0554)
Controls	No	Yes	No	Yes	No	Yes
Cons.	2.680 ***	7.403 ***	2.751 ***	6.025 ***	2.643 ***	0.627 *
	(0.0200)	(1.595)	(0.0214)	(1.192)	(0.0174)	(0.343)
Obs.	1460	1460	1460	1460	1460	1460
R2	0.006	0.046	0.045	0.083	0.083	0.179

Note: VI-PM_2.5, V-PM_2.5, and IV-PM_2.5 represent the analytical findings for PM_2.5 in Haikou, Danzhou, and Sanya, respectively. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

.

The PM_2.5 concentrations in Haikou, Danzhou, and Sanya were measured at 14.433 μg/m³ and 13.668 μg/m³, 15.315 μg/m³ and 13.205 μg/m³, and 12 μg/m³ and 10.83 μg/m³ before and after the implementation of the DECA policy, respectively. Before the incorporation of control variables, the implementation of the DECA policy in Haikou, Danzhou, and Sanya was significantly associated with a reduction in PM_2.5 levels at the 1% significance level. Following the introduction of control variables, the policy continued to demonstrate a significant impact in Danzhou and Sanya, with estimated coefficients of −0.151 ** and −0.148 ***, respectively. In contrast, the estimated coefficient for Haikou was 0.0258. The estimated effects of control variables such as GDP, TFV, and GIO on SO₂ levels in the region were found to be consistent with our previous findings (3.3). Notably, unlike those observed for GBE, the results pertaining to Sanya’s PM_2.5 indicated a significant positive effect at the 1% level.

The findings indicate that the DECA policy is linked to improvements in PM_2.5 pollution levels in Danzhou and Sanya but not in Haikou. Consistent with the work of Zhang et al. (2022), the effectiveness of the DECA policy exhibits heterogeneity [57]. This discrepancy may be attributed to the GIO in Haikou. In 2022, Haikou’s GIO increased by 34% compared to 2021, rising from CNY 574.22 billion to CNY 792 billion. Given that industrial production is a significant contributor to PM_2.5 emissions, the environmental benefits derived from implementing the DECA policy may be insufficient to counterbalance the increase in PM_2.5 resulting from industrial activities. Overall, the estimated results are predominantly negative and statistically significant, suggesting that improvements in PM_2.5 levels within underdeveloped coastal areas are closely associated with this policy initiative. Notably, there exists a positive correlation between GBE and PM_2.5 levels in Sanya. This phenomenon can be explained by a 13.1% rise in total expenditure for Sanya during 2022 alongside a 13.2% reduction in government spending on environmental protection relative to the same period last year; these factors contributed to an elevated GBE index for Sanya and an accompanying increase in PM_2.5 concentrations.

4. Robustness Assessment

In the RD model, bandwidth sensitivity analyses, placebo tests, and doughnut tests are frequently employed as robustness checks for the results [49]. Bandwidth sensitivity analyses were conducted to estimate local treatment effects across varying sample sizes. Generally, utilizing a smaller bandwidth enhances the local precision of estimates; however, it may also increase variance due to a reduced number of data points. Conversely, employing a larger bandwidth can decrease variance but may introduce greater heterogeneity, potentially leading to increased bias in the estimates. Consistent estimation results across different bandwidths can alleviate concerns regarding sensitivity to bandwidth choices and bolster the robustness of the findings. Placebo tests were utilized to ascertain whether the observed discontinuity near the cutoff was attributable to genuine policy changes rather than other endogenous factors. Furthermore, doughnut tests—which exclude observations close to the policy implementation date in order to mitigate manipulation—are employed to examine discontinuities within this adjusted dataset and ensure the robustness of results.

DID models involve two critical tests: the parallel trends test and the placebo test [58]. The primary objective of the parallel trends test is to verify that there are no significant differences in trends between the treatment and control groups prior to policy implementation. If such trend differences exist, they may obscure causal effects, thereby compromising the accuracy of DID estimates. Conversely, the placebo test entails applying the DID model to a time point unrelated to the actual policy intervention before its implementation. This approach aims to determine whether similar causal effects can be observed at this “placebo” time point. If significant effects are detected during this placebo period, it indicates that other factors may influence the DID estimates.

4.1. RD Model–Local Linear Regression Bandwidth Test

In the context of bandwidth testing, local linear regression is a nonparametric regression technique that estimates function values at data points by fitting a linear model within a localized neighborhood surrounding each point. The bandwidth test serves to assess whether the chosen bandwidth in the local linear regression model is appropriate. In this framework, bandwidth controls the size of the neighborhood used for estimation. To mitigate sensitivity to bandwidth in the results, three distinct bandwidths are calculated: the optimal bandwidth, half of the optimal bandwidth, and double the optimal bandwidth. The robustness of the results with respect to different selections of bandwidth was evaluated by re-estimating the RD model test outcomes using these varying bandwidth values. If significant results persist across these different settings, it suggests that the RD estimation findings are robust.

Table 9. Local linear bandwidth test results for SO₂ in inland areas.

Variable	I-SO2	II-SO2	III-SO2
r-Treat	−4.38	−1.59 *	−6.45 ***
	(1.07)	(0.74)	(2.19)
r-0.5Treat	−2.79	−1.77 **	−8.42 ***
	(1.2)	(0.70)	(2.85)
r-2Treat	−3.71	−1.85 *	−3.50 ***
	(0.84)	(0.79)	(1.78)
Obs.	731	731	731
Controls	Yes	Yes	Yes

Note: The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

and

Table 10. Local linear bandwidth test results for results for PM_2.5 in inland areas.

Variable	I-PM2.5	II-PM2.5	III-PM2.5
r-Treat	28.09	−7.423 *	−6.380 *
	(7.757)	(13.31)	(9.762)
r-0.5Treat	−0.785	−12.65 *	−7.909 *
	(5.719)	(7.175)	(6.030)
r-2Treat	5.687	−7.326 **	−4.443
	(4.048)	(6.006)	(5.033)
Obs.	731	731	731
Controls	Yes	Yes	Yes

Note: The significance levels *, and ** correspond to the 10%, and 5% levels, respectively.

present the results of the RD test regarding the effectiveness of the DECA policy on SO₂ and PM_2.5 concentrations in air within inland areas, analyzed under various bandwidths. The RD estimation coefficients exhibit a consistently significant negative correlation, indicating that this policy has a positive effect on urban air quality within the emission control zone. In the analysis of SO₂ levels in Taicang City (Table 9), the optimal bandwidth yields an estimate of −6.45 ***; when halved, it produces −8.42 ***; and when doubled, it results in −3.50 ***—all demonstrating significant negative correlations. This consistency is aligned with our evaluation findings presented in Section 3.1 and Section 3.2. Conversely, while the estimation results for Yibin City (Table 9 and Table 10) generally indicate negative correlations, they remain within a 90% confidence interval. This suggests that cities situated upstream from the Yangtze River’s inland control zone experience minimal effects from the DECA policy concerning SO₂ and PM_2.5 emissions due to limited shipping activities. Furthermore, local linear estimation outcomes for other cities corroborate our empirical findings discussed in Section 3.1 and Section 3.2, revealing improvements in SO₂ and PM_2.5 pollution levels across inland regions—with variations observed based on geographical location. Overall, this conclusion remains robust.

4.2. RD Model–Placebo Test

To assess whether the results of the RD estimation were obtained by chance over time, a series of placebo tests were conducted to enhance the reliability of the findings. This process involved artificially manipulating the breakpoints in the SO₂ and PM_2.5 data by altering the timing of policy implementation. Imaginary breakpoints were established at the 20th, 40th, 60th, and 80th percentiles on both sides of the actual breakpoint, followed by new RD estimations. The results are presented in

Table 11. Placebo test results for SO₂ in inland areas.

Variable	I-SO2		II-SO2		III-SO2
	(1)	(2)	(1)	(2)	(1)	(2)
r-0.2	1.690	−0.130	0.0633	0.946 **	0.570	−1.038
	(2.399)	(1.211)	(0.340)	(0.400)	(2.120)	(4.171)
Obs.	92	91	365	365	365	365
r-0.4	−0.880	1.378	−0.199	0.208	0.478	−0.869
	(1.617)	(2.182)	(0.351)	(0.403)	(2.288)	(1.151)
Obs.	92	91	365	365	365	365
r-0.6	−0.596	3.919	3.224 **	0.0278	1.156	−1.528 ***
	(1.267)	(2.683)	(1.506)	(0.273)	(3.503)	(0.436)
Obs.	92	91	365	365	365	365
r-0.8	−2.062	3.121	0.687	1.011	−3.552	−1.409
	(1.816)	(1.943)	(1.127)	(0.634)	(4.412)	(1.302)
Obs.	92	91	365	365	365	365
Controls	Yes	Yes	Yes	Yes	Yes	Yes

Note: The significance levels **, and *** correspond to the 5%, and 1% levels, respectively.

and

Table 12. Placebo test results for PM_2.5 in inland areas.

Variable	I-PM2.5		II-PM2.5		III-PM2.5
	(1)	(2)	(1)	(2)	(1)	(2)
r-0.2	−1.129	−5.455	1.320	2.590	3.614	2.223
	(6.644)	(10.30)	(5.533)	(9.081)	(3.870)	(9.864)
Obs.	365	365	365	365	365	365
r-0.4	9.626 ***	−4.044	−4.094	−5.877 *	6.529 **	−0.483
	(3.394)	(4.771)	(3.254)	(3.015)	(3.271)	(7.404)
Obs.	365	365	365	365	365	365
r-0.6	−5.899 *	−8.161	−2.750	10.58 ***	−6.149	4.217 *
	(3.147)	(5.603)	(4.612)	(2.479)	(4.587)	(2.311)
Obs.	365	365	365	365	365	365
r-0.8	−22.52	10.07	6.181	−0.990	15.72	4.013
	(21.29)	(11.34)	(10.56)	(9.909)	(16.83)	(4.896)
Obs.	365	365	365	365	365	365

Note: The number of samples on both sides of the original breakpoint multiplied by 0.2, 0.4, 0.6, and 0.8 are used as new breakpoints, represented by r-0.2, r-0.4, r-0.6, and r-0.8, respectively. (1) indicates the left side as the new breakpoint, and (2) indicates the right side. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

. If the estimated coefficients are not statistically significant, this would indirectly suggest that the RD results at the true breakpoint are robust. Conversely, if similar effects are observed after introducing these virtual breakpoints, it may indicate that such results stem from factors other than genuine policy impacts.

Table 11 and Table 12 present the results of placebo tests for the DECA policy’s impact on SO₂ and PM_2.5 levels, respectively, in the inland areas of the three sample cities. The majority of the estimated coefficients do not achieve a significance level of 10%, suggesting that observed differences are not random. This finding indicates that the effect of the policy on air quality was statistically insignificant following alterations to the timing of policy implementation. Furthermore, the positive and negative signs associated with these coefficients reflect a degree of randomness, implying that placebo tests did not uncover consistent trends or directional effects. This stark contrast with the estimated results from genuine breakpoints (3.1 and 3.2) amplifies uncertainty regarding statistical effects after manipulating policy implementation timing. Consequently, this reinforces the robustness of conclusions drawn about the causal relationship between the DECA policy and emission levels of SO₂ and PM_2.5 in inland areas.

4.3. RD Model–Doughnut Test

To ensure that there is no apparent manipulation in the SO₂ and PM_2.5 concentration data surrounding the breakpoint of the RD model, this study employs the “Doughnut RD” method for further robustness analysis. This approach involves re-estimating the model by excluding a segment of the sample around the breakpoint. If the results are consistent with those of the baseline regression, it indicates that there is no evident manipulation of the concentration variables.

Table 13. Donut-hole test results for SO₂ in inland areas.

Variable	I-SO2	II-SO2	III-SO2
r-0.01	−5.139	−3.279 ***	−2.184 ***
	(0.951)	(0.860)	(1.415)
Obs.	183	724	724
r-0.02	−4.861 *	−3.678 ***	−3.659 **
	(1.267)	(1.016)	(1.790)
Obs.	181	716	716
r-0.03	−5.503	−3.675 **	−2.425 *
	(2.230)	(1.630)	(1.594)
Obs.	179	710	710
r-0.04	−4.665 *	−8.676 **	−2.835 *
	(2.390)	(3.899)	(1.556)
Obs.	177	702	702
r-0.05	−3.021	−4.961	−1.360
	(2.961)	(3.490)	(1.729)
Obs.	175	694	694
Controls	Yes	Yes	Yes

Note: The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

and

Table 14. Donut-hole test results for PM_2.5 in inland areas.

Variable	I-PM2.5	II-PM2.5	III-PM2.5
r-0.01	−13.34 *	−32.78 **	−17.56 *
	(7.873)	(12.97)	(9.439)
Obs.	724	724	724
r-0.02	−17.55	−44.28 ***	−33.28 ***
	(11.57)	(12.02)	(9.181)
Obs.	716	716	716
r-0.03	2.410	−26.18	−145.6 ***
	(17.10)	(17.52)	(30.70)
Obs.	710	710	710
r-0.04	−41.13	−58.69 **	−16.66
	(89.61)	(28.51)	(23.05)
Obs.	702	702	702
r-0.05	−237.9	−38.31 *	−2.609
	(198.0)	(32.62)	(32.34)
Obs.	694	694	694
Controls	Yes	Yes	Yes

Note: The symbols r-0.01, r-0.02, r-0.03, r-0.04, and r-0.05 in the table represent the results of re-regression after removing the number of samples on both sides of the breakpoint, multiplied by 0.01, 0.02, 0.03, 0.04, and 0.05, respectively. The significance levels *, **, and *** correspond to the 10%, 5%, and 1% levels, respectively.

present the results of the Doughnut Test concerning DECA policy’s effectiveness in mitigating SO₂ and PM_2.5 pollution in inland areas. The findings indicate that, following the removal of a portion of data surrounding the breakpoint, the RD estimation coefficients for Yibin City are predominantly negative; however, they do not achieve statistical significance at the 90% confidence level. In contrast, for Yichang City and Taicang City, the RD estimation coefficients remain largely negative and generally meet or exceed the threshold for statistical significance at the 90% confidence interval. This outcome effectively rules out potential manipulation and reinforces the robustness of these results.

4.4. DID Model–Parallel Trend Test

The parallel trend assumption inherent in the DID method stipulates that there should be no differential trends between the treatment and control groups prior to the implementation of a policy. To assess this assumption, the dummy variables in Formula (2) are substituted with those corresponding to periods preceding and following the enforcement of the DECA policy. The model employed for testing the parallel trend is as follows (Long and Wan, 2017) [58]:

$$ln{Y}_{it}^{a,b}={{\displaystyle \sum }}_{\tau ϵ\left(-3,-2,-1\right)}{\beta }_{\tau }\left(d{u}_{\tau }\times d{t}_i\right)+\acute{\beta }{Z}_{i\tau }+{\mu }_{\tau }+{\sigma }_i+{\epsilon }_{i\tau }$$

(4)

In the equation, “Time-3,” “Time-2,” and “Time-1” respectively represent the three months before the implementation of DECA, the two months before the implementation, and the month preceding policy enactment, corresponding to the data $d{u}_{-3}$, $d{u}_{-2}$, and $d{u}_{-1} = 1$; otherwise, these values are zero. If the city is within the policy implementation area, then $d{t}_i= 1$; otherwise, the value is zero. The coefficients ${\beta }_{-3}$, ${\beta }_{-2}$, and ${\beta }_{-1}$ are used to evaluate the trend of the policy, and if they are consistent, it indicates the elimination of interference from other policies. ${\mu }_{\tau }$ and ${\sigma }_i$ represent time and country fixed effects, respectively. The other variables are defined in (2).

Table 15. Parallel trends test results for SO₂ in underdeveloped coastal areas.

Variable	IV	V	VI
du−3 × dt	−0.0925	0.0262	−0.0112
	(0.271)	(1.350)	(0.0963)
du−2 × dt	0.146	0.0426	−0.00621
	(0.411)	(1.440)	(0.174)
du−1 × dt	−0.0265	−0.00263	−0.00856
	(0.431)	(1.440)	(0.169)
Controls	Yes	Yes	Yes
Constant	1.985	3.874	−0.687
	(5.171)	(15.21)	(0.589)
Observations	1460	1460	1460
R2	0.754	0.149	0.940

and

Table 16. Parallel trends test results for PM_2.5 in underdeveloped coastal areas.

Variable	IV-PM2.5	V-PM2.5	VI-PM2.5
du−3 × dt	0.0674	−0.127	−0.0346
	(0.712)	(3.819)	(0.389)
du−2 × dt	0.193	−0.0175	−0.0492
	(0.704)	(3.819)	(0.436)
du−1 × dt	0.0106	0.0305	−0.226
	(0.644)	(0.772)	(0.192)
Controls	Yes	Yes	Yes
Constant	17.48 **	8.286	0.182
	(7.606)	(38.10)	(1.381)
Observations	1460	1460	1460
R2	0.230	0.183	0.422

Note: The significance level ** represents the 5% level.

present the results of the parallel-trend tests for SO₂ and PM_2.5, respectively. The findings suggest that the coefficients of $d{u}_{-3}\ast dt$, $d{u}_{-2}\ast dt$, and $d{u}_{-1}\ast dt$ for both the SO₂ and PM_2.5 concentrations are close to zero and not statistically significant at the 90% confidence level. This suggests that there is no systematic difference in the changes in SO₂ and PM_2.5 concentrations between the two groups before policy implementation. In other words, the model satisfied the parallel trends test, providing preliminary evidence for the robustness of the DID model results.

4.5. DID Model–Placebo Test

A placebo test was conducted to eliminate potential unknown factors that could contribute to the overall improvement in SO₂ and PM_2.5 levels. The coefficients of the interaction terms for concentration data were randomly sampled 800 times to assess whether they significantly differed from the baseline regression results (3.3 and 3.4). A combined plot illustrating the coefficients, corresponding p-values, and kernel density of t-values was generated for validation purposes (Figure 4). The blue line indicates the threshold where the p-value is 0.1, while the green line represents the distribution of p-values, and the red line depicts the kernel density distribution of t-values.

The findings indicate that the estimated coefficients of the placebo DID model adhere to a normal distribution. The majority of policy effect coefficients are concentrated around −0.05 to 0.05, closely approaching zero, with most corresponding p-values exceeding 0.1. This lack of significance at the 10% level suggests that, when excluding policy effects, the interaction terms between SO₂ and PM_2.5 did not exhibit significant changes. Consequently, this rules out randomness in the DID estimation results presented in Section 3.3 and Section 3.4. Furthermore, the kernel density of t-values also aligns with a normal distribution without significant deviation from zero, thereby reinforcing the association between improvements in SO₂ and PM_2.5 pollution levels in the three sampled underdeveloped coastal cities and the DECA policy. These findings underscore the robustness of the results reported herein.

5. Discussion

We utilized RD and DID models in econometrics to evaluate the impact of the DECA policy on air quality in both inland and coastal underdeveloped cities [16,57,59]. The selected sample cities included Yibin, Yichang, and Taicang, which are situated in various segments of the inland DECA region, as well as the underdeveloped coastal cities of Haikou, Danzhou, and Sanya.

According to the analysis of the inland region, the policy effects in Yibin, located upstream of the DECA region, did not exceed the 90% confidence interval. This finding suggests that the DECA policy has not yielded positive outcomes in mitigating SO₂ and PM_2.5 pollution in that area [60]. In contrast, for Yichang and Taicang, situated in the middle and lower reaches of the river, the estimated coefficients for SO₂ concentration achieved a significance level of 5%. Similarly, for PM_2.5 concentrations, the estimated coefficients reached significance levels of 5% and 10%, respectively. Among them, the average daily concentration of SO₂ exhibited a decrease of 5.3% and 18.9%, respectively, while the average daily concentration of PM_2.5 showed reductions of 21.9% and 13.9%, respectively. Furthermore, these estimation results have successfully passed three robustness tests, thereby confirming their reliability. In analyzing underdeveloped coastal cities, it was found that the DECA policy significantly reduced SO₂ concentrations in ambient air [61]. Cities such as Haikou, Danzhou, and Sanya exhibited significant improvements at a 1% significance level, and the average daily concentration of SO₂ decreased by 1.6%, 20.2%, and 21.2%, respectively. A study examining the reduction of PM_2.5 concentrations in coastal underdeveloped regions as a result of DECA policies revealed that PM_2.5 levels in Danzhou and Sanya experienced significant decreases of 13.8% and 9.5%, respectively, with a significance level of 1%. The effect of this policy on Haikou was significant prior to incorporating control variables; however, after their inclusion, it failed to meet the criteria for passing through a 90% confidence interval. This observation implies that increased industrial output may be an influencing factor. The results derived from the DID model have successfully passed both parallel trend and placebo tests—further indicating their robustness. Therefore, it can be concluded that the DECA policy has produced positive effects on improving air quality across both inland regions and underdeveloped coastal areas [62].

Despite the overall improvement in air quality attributed to the DECA policy in both inland and coastal underdeveloped regions, its effectiveness exhibits variability across different cities. In inland cities, most control variables within the RD estimation results did not achieve significance at the 90% confidence level, suggesting a limited impact of these variable indices on gas pollutant concentrations. This indicates that the variation in these economic indicators exerts a limited direct influence on the concentration of gas pollutants addressed by the DECA policy in these inland regions. We propose that this notable contrast with the significant findings observed in coastal cities arises from fundamental differences in pollution source composition and economic structure.

The primary sources of pollution in the inland cities examined (Yibin, Yichang, Taicang) are likely dominated by local stationary sources, including heavy industry, manufacturing activities, and coal combustion for residential heating and power generation [63]. While economic activity—reflected through metrics such as GDP and GIO—certainly contributes to the baseline levels of pollution, its short-term fluctuations may be obscured by the substantial and relatively stable contributions from these predominant local sources. Moreover, the DECA policy specifically targets mobile emissions from shipping; however, this sector constitutes a negligible fraction of total emissions in these inland river ports when compared to their industrial base. Consequently, the effectiveness of this policy in these regions is more influenced by geographical factors that determine the intensity of shipping activity (e.g., location along the river affecting vessel traffic density), rather than short-term variations in broader economic indicators. This observation elucidates the pronounced heterogeneity based on geographic location: significant effects are noted primarily in the middle and lower reaches (Yichang, Taicang), where shipping volumes are higher; conversely, no notable effects were observed in the upstream city (Yibin), which experiences lower traffic levels.

Conversely, the economies of underdeveloped coastal cities such as Haikou, Danzhou, and Sanya are inherently more intertwined with maritime activities. Port operations, logistics, and tourism—all of which are highly dependent on shipping—constitute a critical component of their economic structure. Consequently, variables such as GDP, TFV (Total Freight Volume), and GIO (Gross Income from Operations) exhibit a more direct correlation with ship traffic volume and, consequently, with the emissions regulated by the DECA policy [64]. An increase in these economic indicators directly correlates with heightened shipping activity, thereby rendering the reductions achieved through low-sulfur policies more pronounced and statistically detectable against this baseline. The observed heterogeneity can thus be logically attributed to variations in city-specific development indices; these indices serve as proxies for the scale of pollution sources associated with shipping activities.

The DECA policy has demonstrated varying degrees of positive impacts on air quality in both inland and underdeveloped coastal cities. Nevertheless, despite the implementation of regulations aimed at reducing emissions, it is estimated that ship emissions contribute to approximately 250,000 premature deaths globally each year [65]. Currently, under China’s DECA policy, vessels operating in Hainan Province are mandated to adhere to a sulfur content limit of 0.1% (m/m). For other coastal waters, it remains undetermined whether this limit will be reduced from 0.5% to 0.1% (m/m) by the year 2025. In light of these research findings, it is recommended that all coastal waters implement a sulfur content limit policy set at 0.1% (m/m) in order to mitigate ship emissions and associated pollution. At the same time, to effectively mitigate the pollution caused by ship emissions, it is essential to closely monitor whether the metal content in the particulate matter discharged from ships complies with the relevant standards [66].

This study acknowledges several limitations while also highlighting opportunities for future research. First, a primary limitation is the discrepancy in data frequency among core variables. The integration of daily average concentrations of SO₂ and PM_2.5 with low-frequency macroeconomic data can elucidate significant long-term trends; however, it may not accurately reflect the immediate effects of short-term fluctuations in economic activity on air quality. This mismatch could result in measurement errors when estimating the influence coefficients of these variables. To address this limitation, future research should consider employing higher-frequency proxy variables to achieve better alignment with environmental data. Utilizing such high-frequency surrogate variables is anticipated to provide a more precise quantification of the short-term dynamic relationship between macroeconomic activities and environmental pollution, thereby validating and extending the conclusions drawn from this study.

This research is based on public data, which not only ensures the transparency of the entire research process but also facilitates the subsequent expansion of related fields. Currently, ship emissions and air quality data are scattered between maritime departments, environmental agencies, and businesses, creating “data silos” that seriously hinder accurate assessment of control zone policies. It is recommended to establish a cross-departmental unified data warehouse and advocate for the public release of de-identified high-precision data in accordance with the principles of findable, accessible, interoperable, and reusable (FAIR). This initiative will greatly improve the transparency of policy research and provide an important data basis for the dynamic optimization of future policies.

6. Conclusions

Ship emissions not only impact air quality and human health in developed port cities but also affect inland and underdeveloped coastal areas. This study employs econometric methods to evaluate the implementation effects of China’s ECA Policy, enacted in 2018, specifically within both inland and underdeveloped coastal cities. The policy’s effectiveness in mitigating ship emissions for these regions was analyzed separately using RD and DID models.

Our findings indicate that the DECA policy has made a positive contribution to the enhancement of air quality in both inland and underdeveloped coastal regions, aligning with previous research. However, this study uncovered heterogeneity in the policy’s effectiveness regarding air quality improvement, which is influenced by various geographical locations and developmental statuses of the cities. This variability indicates that the effectiveness of the policy is not universally applicable but rather influenced by local factors. In inland cities, the geographical position along the river emerged as the primary determinant of success. Conversely, in coastal cities, socioeconomic development indices served as key moderating factors, likely due to their stronger correlation with the scale of local shipping activity.

These findings underscore the multifaceted and complex impacts of ECA policies, highlighting the necessity for a comprehensive assessment of ECA effectiveness. This study addresses the gap in evaluating China’s initiatives to regulate ship emissions by examining the implementation of ECA policy in both inland and underdeveloped coastal cities. It offers valuable insights for future comprehensive evaluations of the environmental benefits associated with ECA policies and serves as a reference for other countries seeking to develop similar policies within the region. More importantly, this study provides an important evidence base for designing sustainable maritime policies that both safeguard public health and promote economic development, thereby contributing to the sustainable development goals. The conclusions presented above are derived from macroeconomic low-frequency data. However, the dynamic interactions on shorter time scales require further validation through the application of higher-frequency data in future research.