Causality and dynamic spillover effects of megacities on regional industrial pollution reduction

Abstract

Regional economic power and local environmental policies have a substantial impact on pollution reduction in urban agglomerations (UAs); however, whether megacities in UAs exert spillover effects of pollution reduction on surrounding cities remains unknown. This study presents a causal analytic framework to evaluate the spillover effects of megacities on regional industrial pollution reduction in three major UAs in China between 2005 and 2016. The interaction between industrial pollution reduction and infrastructure investment indicators was also examined. Results indicated a good fit for spatial spillover of sulfur dioxide reduction (SR) in the Pearl River Delta (PRD) and Yangtze River Delta (YRD) but not in the Beijing-Hebei-Tianjin cluster (JJJ). Spatial spillover of dust reduction (DR) was evident in the PRD and JJJ but not the YRD. Spatial analysis showed that infrastructure investment indicators, at megacity and UA levels, had short-term spillover effects on surrounding cities for DR but not SR. However, spatial spillover effects, at both the city and UA levels, were substantial over the long term. In addition, the results of the spatial-time lag analysis suggest a linear relationship between pollution control-related infrastructure investment indicators and long-term pollution reduction. This study provides new information regarding the spatial spillover effects of megacities on regional industrial pollution reduction in UAs.

Highlights

• •This study investigated the spillover effect of UA megacities on pollution reduction.
• •Interaction between pollution reduction and infrastructure investment was examined.
• •RelAtionship between infrastructure investment and pollution reduction was linear.
• •Spillover effects at both city and UA levels were substantial in the long-term.

1. Introduction

Urbanization has caused many ecological and environmental issues worldwide, particularly in areas undergoing rapid infrastructure investment. At the frontline of urban agglomeration (UA), megacities (clusters of cities with populations of more than 10 million) face serious industrial pollution problems that threaten human health and cause social and economic losses [1]. Therefore, understanding the relationship between the development of UAs and environmental pollution is crucial to regional sustainability. Extant studies have examined the environmental pollution effects of UA, with mixed results. Some studies have found that UAs exacerbate environmental pollution [2,3], whereas others maintain that UAs encourage enterprises to adopt environmentally friendly technologies and thus, reduce environmental pollution [[4], [5], [6]]. Other studies have found that the relationship between UA development and environmental pollution is uncertain or nonlinear [7]. The development of urban and industrial agglomerations improves industrial efficiencies and plays a positive role in the economies of scale; however, it also produces increasingly serious negative environmental externalities. For example, the Beijing-Hebei-Tianjin (JJJ) cluster and other regions of China experience several heavy haze days every year. The development of urban and industrial agglomerations has reduced environmental quality and polluted the ecological environment. However, based on the externality approach [8], such development can also promote energy conservation and emission reduction, thereby improving regional environmental quality. For example, local governments can implement unified and effective environmental regulation policies for enterprises in industrial parks, which will produce a scale effect of environmental governance. Therefore, it is necessary to re-examine the relationship between urbanization and environmental pollution and explore the mechanisms of spatial spillover effects of pollution control in UAs.

In recent years, with the promotion of China's ecological civilization construction, regional pollution emission control has increasingly become a research focus. Infrastructure investment supposedly strengthens pollution control to some extent [9]. Compared with other cities, the government invests more in the infrastructure of megacities with a focus on pollution reduction, enabling the infrastructure of megacities to play an exemplary role in economic development and pollution control, as well as in driving spillover effects on the surrounding cities [10,11]. However, infrastructure investment by Chinese megacities in UAs has shown different trends in the past few years, leading to significant differences in the efficiency of pollution reduction [12]. Whether megacities can take advantage of the positive spatial spillover effects that play a leading role in pollution reduction of the cities surrounding UAs is an important focus area for research. This study aims to explore the spatial spillover effect of infrastructure investment in megacities and whether megacities can take advantage of this effect within UAs to reduce industrial pollution.

1.1. Spatial spillover effect of UAs and magacities

Differences in resource endowments and economic levels are comparative advantages for cities [13]. [14] adopted the difference-in-differences model and found that urbanization and the construction of a National New District promoted economic development but failed to effectively reduce air pollution emissions. If spatial spillover effects are considered, then UAs can reduce the per capita smoke emission of treated cities and therefore, can promote economic development and reduce air pollution [15]. Cities can use spatial spillover effects of agglomerated economies to improve the overall sustainable development of UAs [16,17]. In addition, the spatial agglomeration of industries produces economies of scale and reduces environmental pollution [4,6,18,19]. Other studies suggest that due to industrial agglomeration and the concentration of production activities, urban enterprises would increase their energy demand when agglomerated [2,20], which would further increase energy consumption and pollutant emissions. The dynamic relationship between urban concentration and environmental pollution has been examined and it was found that UAs had a strong positive impact on air pollution emissions in most sample countries [21]. This is due to high population densities, overcrowding, traffic congestions, and high energy consumption demands of big cities. However, in some countries, the relationship between UAs and pollution emissions has shown an overall downward trend.

Megacities play a leading role in the development of UAs and have a great impact on the spillover of industrial pollution within the region. Some studies have focused on the impact of environmental regulations on air pollution reduction [22,23]. [24] elaborated on the impact of government regulation on corporate environmental performance. Other studies claimed that air pollution is affected not only by local environmental regulations but also by regulations implemented by surrounding cities [23,25]. Cities with lax regulations become pollution havens and spatial spillovers offset local environmental regulations focused on improving air quality. By comparing urban agglomerations, they found that the spatial spillover effects of dust pollution in JJJ and Yangtze River Delta (YRD) were higher than those of Pearl River Delta (PRD). Therefore, spatial spillover effects should be considered, and joint regulations should be strengthened to deal with air pollution in UAs. Sulfur dioxide pollution causes great damage to urban air quality and the ecological environment. It also aggravates the formation and pollution of fine particulate matter (PM_2.5). Therefore, there is considerable scope for further study of spatial spillover of SO₂. In particular, it is necessary to explore the spatial spillover effect of SO₂ in megacities.

1.2. Influence of infrastructure investment on industrial pollution reduction

Existing studies have proven that increasing investment in urban transportation infrastructure can alleviate air pollution on the whole, but long- and short-term effects are significantly different. The proportion of investment for infrastructure is much larger in megacities than in other types of cities and it is driven by massive economic growth and infrastructure needs, as well as government policies, in rapidly urbanizing countries. Studies have demonstrated a significant effect of environmental management and infrastructure investment on reducing pollution in megacities [[26], [27], [28]] and found an inverted U-shaped relationship between infrastructure investment and industrial dust pollution in UAs [29].

Extant research has explored the spatial spillover effects of infrastructure investment in megacities on industrial pollution reduction in UAs. However, this is the first study to investigate the spatial spillover effects of infrastructure investment in megacities, and whether these megacities take advantage of spatial spillover effects within UAs for industrial pollution reduction. This study contributes to the literature by providing complementary explanations for industrial production control in UAs. Since infrastructure investments today will determine pollution emissions in the future; the long term spatial spillover effects of pollution reduction and investment need to be analyzed. By combining infrastructure investment indicators with pollution reduction, this study attempts to construct a causal analytic framework to evaluate the short- and long-term spillover effects of infrastructure investment in megacities on SO₂ and dust pollution reduction in surrounding cities.

2. Materials and methods

This study hypothesized that infrastructure investment in megacities has a spatial spillover effect on surrounding cities and that megacities can use the spatial spillover effect within UAs to reduce industrial pollution. The study used ANOVA and the bootstrap autoregressive-distributed lag (ARDL) test, with SR and DR as the dependent variables. The time series of pollution control-related infrastructure investment indicators gross domestic product (GDP), gross domestic product per capita (GC), fixed asset investment (FI), and fixed asset investment per capita (FIC) were the independent variables. Spatial-time lag analysis can confirm both short- and long-term (time series with rolling correlation) spillover effects from megacities in the three major Chinese UAs of this study. The long-term causality among infrastructure investment indicators was examined for megacities, along with pollution reduction outcomes, using data from 2005 to 2016 for the three UAs.

2.1. Empirical framework

ANOVA was used to investigate DR and SR spatial spillover effects of megacities in the PRD, YRD, and JJJ on a short-term temporal scale, whereas the bootstrap ARDL model was used to test the long-term spillover effects at both city (megacity vs. other cities) and UA levels (with and without megacities). The bootstrap ARDL test, a rigorous method to test the long-term co-integration relationship between time series variables, simulates the cointegration relationship between variables when setting a single equation time series [30]. More importantly, this model can further test for short-term causality. Traditionally, an ARDL test determines the direction of short-term causality based on the Granger causality test [31]. If y is caused by a variable, there is no cointegration between y and x. Therefore, a Granger test of x→y should only include the lag difference of x. Compared with the traditional ARDL model, the bootstrap ARDL can more rigorously test co-integration. First, the traditional ARDL model assumes that two variables cannot be exchanged into dependent and independent variables, whereas the bootstrap ARDL test allows two or more variables to be exchanged into dependent and independent variables. Second, a bootstrap ARDL cointegration test was proposed by applying the bootstrap method to the ARDL cointegration test. This approach has the advantage of being able to test variables with the proper sample size and power properties. Finally, bootstrap ARDL improves the rigor of a traditional ARDL test as it adds explanatory variables of backwardness, thereby complementing the F and T tests of the traditional ARDL co-integration test. Pre-test variables are not required while using a bootstrap ARDL test [30]. Regardless of the integration order of variables, this method can be used if stationarity can be achieved at or below the first difference and can be used to calculate short- and long-term estimates simultaneously. The bootstrap ARDL modeling equations used in this study are listed (Table 1).

Table 1.

Bootstrap ARDL modeling equations.

${Y_{DR,cl}}_t={\alpha }_{DR,cl}+{\beta }_{DR,cl}{X}_{{GC,cl}_t}^i+{\gamma }_{DR,cl}{Z}_{{FIC,cl}_t}^j+{\epsilon }_t(1)$
${Y_{DR,ual}}_t={\alpha }_{DR,ual}+{\beta }_{DR,ual}{X}_{{GC,ual}_t}^i+{\gamma }_{DR,ual}{Z}_{{FIC,ual}_t}^j+{\epsilon }_t(2)$
${Y_{SR,cl}}_t={\alpha }_{SR,cl}+{\beta }_{SR,cl}{X}_{{GC,cl}_t}^i+{\gamma }_{SR,cl}{Z}_{{FIC,cl}_t}^j+{\epsilon }_t(3)$
${Y_{SR,ual}}_t={\alpha }_{SR,ual}+{\beta }_{SR,ual}{X}_{{GC,ual}_t}^i+{\gamma }_{SR,ual}{Z}_{{FIC,ual}_t}^j+{\epsilon }_t(4)$

Note: $cl$: city level effects; $ual$: UA level effects; GC: GDP per capita; FIC: fixed asset investment per capita.

The error correction version of the bootstrap ARDL model is shown below:

$${\Delta y_t=\alpha }_0+{\sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathrm{\alpha }}_{\mathrm{i}}\Delta y}_{\mathrm{t}-\mathrm{i}}+{\sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathrm{\beta }}_{\mathrm{i}}\Delta x}_{\mathrm{t}-\mathrm{i}}+{\sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathrm{\gamma }}_{\mathrm{i}}\Delta z}_{\mathrm{t}-\mathrm{i}}+{\lambda }_s+u_t$$ (5)

$${{\lambda }_s=\lambda }_1{y}_{t-1}+{\lambda }_2{x}_{t-1}+{\lambda }_3{z}_{t-1}$$ (6)

where ${\alpha }_0$ is a constant parameter; ${\mathrm{\alpha }}_{\mathrm{i}},{\mathrm{\beta }}_i$, and ${\mathrm{\gamma }}_{\mathrm{i}}$ are the short-term dynamics of the model; ${\lambda }_s$ is the long-term relationship; $u_t$ is the stationary white noise process for outcome $Y$; and ${\lambda }_1+{\lambda }_2+{\lambda }_3=0$ is the null hypothesis in the ARDL test that represents the nonexistence of a long-term relationship.

The model stationarity was checked by testing the ACF of each variable. For a stationary signal, the ACF values are expected to be zero for each time lag (τ) because no dependence on the factor of time is expected. The correlation for time series observations was calculated using observations from previous time steps (lags).

2.2. Data sources

Data were obtained from multiple sources, encompassing industrial dust production, industrial dust emissions, industrial SO₂ production, industrial SO₂ emissions, GDP, GC, FI, and FIC. The constructs, unit calculation methods, and data sources are listed in Table 2. The dependent variables in the ANOVA and bootstrap ARDL tests, including the DR and SR percentages, were calculated from data on industrial dust production, industrial SO₂ production, industrial dust emissions, and industrial SO₂ emissions. The infrastructure investment indicators included GC and FIC and were selected as independent variables for the bootstrap ARDL test. The cities within the PRD, YRD, and JJJ are presented in the Appendix (Table A1).

Table 2.

Variables and data sources.

Variable	Unit/calculation method	Data Source
Industrial dust production	Tons	CCSY
Industrial SO2 production	Tons	CCSY
Industrial dust emission	Tons	CCSY
Industrial SO2 emission	Tons	CCSY
Gross domestic product	100 million RMB	CCSY
Fixed asset investment	100 million RMB	CCSY
Urban agglomerations	YRD, PRD, JJJ	CCC and SCC

Note: YRD: Yangzi River Delta; JJJ: Beijing-Hebei-Tianjin cluster; PRD: Pearl River Delta; CCSY: China City Statistical Yearbook; CCC: CPC Central Committee; SCC: State Council of China.

In the current study, the long-term causality relationship between industrial pollution reduction and infrastructure investment indicators was assessed for spatial and temporal effects using a bootstrap ARDL test. The spatial effects were measured at the city and UA levels. The time effect was measured by testing the time series data from 2005 to 2016.

The ANOVA and bootstrap ARDL tests were performed using R (www.r-project.org). The R package “dLagM” was used for the bootstrap ARDL test27. The descriptive results of the variables are provided (Table 3). The count (108) is the number of rows in the data, which includes panel data of three UAs at three levels (Municipals, Vice-province, and Prefecture) in the past 12 years (n = 12 years × 3 UAs × 3 levels = 108). The median of each variable is (FI: 1.68E+07; FIC: 3.25; GDP: 3.88E+07; GC: 1.21E+05; SR: 0.64; DR: 0.99). The kurtosis of the variables indicate that FI, GDP, GC, and DR have heavier tails (ourliers) than FIC and SR, while the skewness shows that FI, FIC, GDP, AND GC are skewed right, whereas SR and DR are skewed left. FIC and SR are relatively close to the normal distribution.

Table 3.

Descriptive results of the variables.

	FI	FIC	GDP	GC	SR	DR
Mean	2.26E+07	3.46	6.29E+07	4.15E+05	0.62	0.98
Standard Error	1.67E+06	0.15	5.46E+06	6.07E+04	0.02	0
Median	1.68E+07	3.25	3.88E+07	1.21E+05	0.64	0.99
Standard Deviation	1.74E+07	1.51	5.68E+07	6.31E+05	0.17	0.01
Sample Variance	3.03E+14	2.28	3.23E+15	3.98E+11	0.03	0
Kurtosis	2.35	−0.18	1.57	2.44	0.24	2.27
Skewness	1.54	0.47	1.5	1.89	−0.69	−1.18
Range	8.09E+07	6.85	2.55E+08	2.50E+06	0.83	0.05
Minimum	3.80E+06	0.66	9.88E+06	1.58E+04	0.12	0.95
Maximum	8.47E+07	7.51	2.65E+08	2.51E+06	0.95	1
Sum	2.44E+09	373.15	6.80E+09	4.48E+07	66.68	106.28
Count	108	108	108	108	108	108
Confidence Level (95.0%)	3.32E+06	0.29	1.08E+07	1.20E+05	0.03	0

3. Results

We summarized the basic characteristics of three UAs: the PRD, YRD, and JJJ, and compared their values of the pollution control-related infrastructure investment indicators (GDP, GC, FI, and FIC) (Table 4).

Table 4.

Basic characteristics of the three urban agglomerations (UAs) in China.

Urban agglomeration	Abbreviation	Year of approval	Population (million)	Regional GDP (trillion dollars)
Beijing-Hebei-Tianjin cluster	JJJ	2015	112.7	1.33
Pearl River Delta	PRD	2015	57.2	1.54
Yangzi River Delta	YRD	2016	225	1.14

The regional GDP and FI of all three UAs, maintained a strong upward momentum (Fig. 1, Fig. 2, Fig. 3, Fig. 4). As important carriers of regional development, UAs face severe environmental pollution problems [32]. Owing to limited data in the China City Statistical Yearbook (CCSY), the current study focused on the reduction of SO₂ and dust from industrial pollution, which can be easily distinguished from other domestic pollutants.

Fig. 1.

Total GDP of the three major UAs in China. YRD: Yangzi River Delta; JJJ: Beijing-Hebei-Tianjin cluster; and PRD: Pearl River Delta.

Fig. 2.

GDP per-capita of the three major UAs in China. YRD: Yangzi River Delta; JJJ: Beijing-Hebei-Tianjin cluster; and PRD: Pearl River Delta.

Fig. 3.

Total fixed asset investments and fixed asset investments per-capita of the three major UAs in China. YRD: Yangzi River Delta; JJJ: Beijing-Hebei-Tianjin cluster; and PRD: Pearl River Delta.

Fig. 4.

Fixed asset investments per-capita of the three major UAs in China. YRD: Yangzi River Delta; JJJ: Beijing-Hebei-Tianjin cluster; and PRD: Pearl River Delta.

The SR and DR efficiencies in the PRD, YRD, and JJJ are shown in Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9, Fig. 10. The development status of each UA reflected their development trends and problems. There was a great imbalance in the development of all three UAs, which was reflected by the large gap in UA developments in the hinterland, especially in JJJ. These UAs have different development trajectories. The north region of China, where JJJ is located, has abundant natural resources such as coal, iron ore, and oil. It has traditionally been the major heavy industrial base of China and contributes largely to air pollution [33]. The YRD far exceeds the other two UAs in the context of tertiary industry FI and GDP [34]. The PRD is at the forefront of China's reform and has had many achievements in the construction of population, economy, life, culture, and urban systems. The pollution reduction effect of tertiary industry agglomeration is increasing. As a product of industrialization and urbanization, megacities have become sources of the economic radiation of UAs owing to their huge populations, economic scales, and strong aggregation of resources [35].

Fig. 5.

SO₂ reduction efficiency in the Pearl River Delta (PRD).

Fig. 6.

Dust reduction efficiency in the Pearl River Delta (PRD).

Fig. 7.

SO₂ reduction efficiency in the Yangtze River Delta (YRD).

Fig. 8.

Dust reduction efficiency in the Yangtze River Delta (YRD).

Fig. 9.

SO₂ and dust reduction efficiency in Beijing-Hebei-Tianjin (JJJ).

Fig. 10.

Dust reduction efficiency in Beijing-Hebei-Tianjin (JJJ).

Pearson's correlation coefficient, which describes either a negative or positive correlation is shown (Fig. 11). The autocorrelation function (ACF) values of SR, at city and UA levels, were consistent for the long term. Similarly, the DR (at city and UA levels) lines also basically coincided in the long term. The long-term trends in SR and DR tended to be zero, which confirmed the autocorrelation and stationarity of these two variables.

Fig. 11.

Autocorrelation functions (ACF) of SO₂ reduction (SR) and dust reduction (DR).

The correlation matrix (Table 5) indicates that FI, FIC, GDP, and GC are significantly correlated, except for the correlation between FIC and GC. The SR was significantly correlated with all variables, with the GC correlation being the most significant. DR also showed significant correlations with all variables, except for GC. The results showed that the economic power of megacities (represented by GC) significantly influenced SR, whereas FI was more important for DR. Meanwhile, FIC had the highest correlations with both SR and DR, which shows that they were both significantly influenced by FIC.

Table 5.

Correlation matrix of variables.

	FI	FIC	GDP	GC	SR	DR
FI	1
FIC	0.59***	1
GDP	0.86***	0.54***	1
GC	−0.12	0.21*	−0.20*	1
SR	0.28*	0.67***	0.28**	0.45***	1
DR	0.36***	0.48***	0.36***	0.18	0.47***	1

Notes: n = 108; Significance: ‘***’ 0.001; ‘**’ 0.01; ‘*’ 0.05; GDP: gross domestic product; GC: GDP per capita; FIC: fixed asset investments per capita; FI: fixed asset investments; SR: SO₂ reduction; and DR: dust reduction.

The goodness-of-fit from the ANOVA test is presented (Table 6). The SR and DR fitness values, at the city and UA levels, are presented in the Appendix (Figures A1–A4). The ANOVA test was significant at the 5% level and showed a good fit for the ${\chi }^2$ value of the spatial spillover effects of GC and FIC on DR, at the city level (spatial: ${\chi }^2$ = 30.161, p < 0.001; temporal: ${\chi }^2$ = 41.187, p < 0.001). The spatial and temporal effects of GC and FIC on DR at the UA level were also significant (spatial: ${\chi }^2$ = 30.451, p < 0.001; temporal: ${\chi }^2$ = 41.583, p < 0.001). These results indicate that the infrastructure investment indicators, at both city and UA levels, had DR spillover effects on the surrounding cities. The results of GC and FIC for SR were more complicated. The time effects of GC and FIC on DR, at the city level (${\chi }^2$ = 164.3404, p < 0.001) and UA level (${\chi }^2$ = 170.9309, p < 0.001), were significant. However, the spatial effects were not significant. Therefore, infrastructure investment indicators, at the megacity and UA levels, did not have short-term SR spillover effects on the surrounding cities. The cost of reducing SO₂ emissions was higher than that of DR. Surrounding cities need to invest more money and time to implement SR measures. Therefore, the spatial spillover effect of SR was not significant in the short term.

Table 6.

ANOVA results of GC and FIC on DR at the city level.

	City Level (spatial)	City Level (time)	UA Level (spatial)	UA Level (time)
DR	30.161***	41.187***	30.451***	41.583***
SR	1.1290	164.3404***	1.1742	170.9309***

Note: Significance: ‘***’ 0.001; ‘**’ 0.01; ‘*’ 0.05.

The goodness-of-fit measure and significance of ANOVA for evaluating the statistical differences between industrial pollution reduction, GC, and FIC in the three UAs are reported (Table 7, Table 8). The ANOVA test was significant at the 5% level (Table 7) and showed a good fit for the ${\chi }^2$ value on the spatial spillover of SR in the PRD (${\chi }^2$ = 0.36677, p < 0.01) and YRD (${\chi }^2$ = 0.349158, p < 0.01) but not in JJJ. The spatial spillover of DR was evident in the PRD (${\chi }^2$ = 0.002611, p < 0.001) and JJJ (${\chi }^2$ = 0.000273, p < 0.05), but not in the YRD. The F-statistic values of SR and DR are shown (Table 8). The F-values of GC and FIC on SR in the PRD (F = 7.650474), YRD (F = 6.540209), and JJJ (F = 0.097899) highlighted that the long-term effects in the PRD and YRD were stronger than that in JJJ. The F-values for GC and FIC on DR in the PRD (F = 14.61438), YRD (F = 0.160158), and JJJ (F = 4.020002) showed that the spatial spillover effect in the PRD was stronger than that in JJJ, whereas the spatial spillover effect in the YRD was the least pronounced. These three city clusters have distinct industrial structures [36]. The YRD and PRD focus on light industry, whereas JJJ focuses on heavy industry, which tends to produce more SO₂ pollution [13]. Tianjin's leading role in JJJ and SR spillover effects were not fully demonstrated due to its industrial structure and geographical location. Environmental regulations play a crucial role in pollution reduction as well as spatial spillover effects in UAs. In recent years, the Chinese government has been committed to controlling dust pollution and has achieved remarkable results, as most megacities have played a leading role in dust pollution reduction [25]. There are three megacities in the PRD (Shenzhen, Guangzhou and Dongguan), and the spatial spillover effect of the PRD is stronger than other two UAs. It also confirms the spillover effect of megacities on pollution reduction of the entire UA. Although the PRD and YRD are dominated by light industries, the spillover effects of DR differ. In the eastern coastal region of the YRD, dust pollution is not severe. In the other cities that surround megacities in the YRD, there have been lower levels of dust pollution, resulting in insignificant DR spillover effects [37]. Due to the industrial characteristics and geographical location, the dust pollution of the YRD is lower than that of other UAs. Therefore, DR spillover effects were not significant in the YRD.

Table 7.

Spatial spillover of megacities in the Pearl River Delta (PRD), Yangtze River Delta (YRD), and Beijing-Hebei-Tianjin (JJJ).

UA	SR	DR
PRD	0.36677**	0.002611***
YRD	0.349158**	1.85E-05
JJJ	0.004489	0.000273*

Note: Significance: ‘***’ 0.001; ‘**’ 0.01; ‘*’ 0.05.

Table 8.

F-statistic values of SR and DR.

UA	SR	DR
PRD	7.650474**	14.61438***
YRD	6.540209**	0.160158
JJJ	0.097899	4.020002*

Note: Significance: ‘***’ 0.001; ‘**’ 0.01; ‘*’ 0.05.

Comparing the F-values from different time lags (i and j = 1,2,3) in the bootstrap ARDL model, the best results were obtained when i and j = 1 (Table 9). The results showed that the infrastructure investment indicators of megacity (F = 16.49, p < 0.001) and UA levels (F = 22.92, p < 0.001) had long-term spillover effects of DR on the surrounding cities (Table 10). The infrastructure investment indicators of megacity (F = 75.52, p < 0.001) and UA levels (F = 107.5, p < 0.001) had long-term spillover effects of SR on the surrounding cities. Studies in this field have found linear [38], U-shaped [39], or N-shaped relationships [40] between economic and demographic factors and pollution reduction outcomes. The results from the current study suggest a linear relationship between infrastructure investment indicators and long-term pollution reduction. This is because China was already in the period of economic take-off from 2005 to 2016, and many UAs and megacities paid more attention to environmental pollution controls. Therefore, contrary to U-shaped or N-shaped relationships or inverted U relationships determined by previous studies, the results of the current study indicate that with an increase in infrastructure investment, there was a positive spatial spillover effect of megacities on pollution reduction on surrounding cities. In the long-term, pollution control-related infrastructure investment in megacities can effectively reduce industrial pollution and have a positive radiation effect on other cities in UAs. The spatial-time differences in SR and DR are presented in the Appendix (Figures A5–A6).

Table 9.

Comparison of the F-statistics from bootstrap ARDL.

	DR (City Level)	DR (UA Level)	SR (City Level)	SR (UA Level)
i = 1, j = 1	16.49***	22.92***	75.52***	107.5***
i = 2, j = 2	11.97***	14.91***	69.33***	83.03***
i = 3, j = 3	14.22***	18.78***	66.71***	90.52***

Note: Significance: ‘***’ 0.001. The bold numbers are the best results of comparison of F-statistics of bootstrap ARDL.

Table 10.

Results of spatial-time lag analysis.

	City Level	UA Level
DR	16.49*** (i = 1, j = 1)	22.92*** (i = 1, j = 1)
SR	75.52*** (i = 1, j = 1)	107.5*** (i = 1, j = 1)

Note: Significance: ‘***’ 0.001; ‘**’ 0.01; ‘*’ 0.05.

4. Discussion and environmental implications

The current study investigated the short- and long-term causality among GC, FC, SR, and DR in three UAs in China from 2005 to 2016. The ANOVA results indicate that the spatial spillover effects of GC and FIC on DR, at both city and UA levels, were significant, whereas the spatial effects of GC and FIC on SR were insignificant. There are several reasons for this result. The pollution transfer effect of certain megacities greatly reduces their role and technological driving effects on other cities [41]. Reductions in primary and secondary pollution emissions in the target area occur at the cost of increased emissions from nearby cities [42,43]. The cost of reducing SO₂ emissions is high. Surrounding cities need to invest more money and time to implement pollution reduction measures from frontier cities [44]. Therefore, the spatial spillover effects of SR were insignificant compared to those of DR in the short term. In recent years, the Chinese government has increased its commitment to controlling urban dust pollution [45]. This makes the spatial spillover effect of DR significant at both city and UA levels.

ANOVA analysis also revealed a good fit for the spatial spillover of SR in the PRD and YRD, but not in JJJ. The three UAs have distinct industrial structures. The JJJ region focuses on heavy industries, whereas the YRD and PRD areas focus on the development of light industries. The development of heavy industry generates more SO₂. Therefore, the leading role and SR spillover effect of the megacity (Tianjin) in the JJJ region did not provide evidence for support. In addition, owing to climatic and geographical locations, coastal cities in the YRD and PRD are conducive to the diffusion of SO₂ pollutants. The SO₂ concentration is therefore relatively low [23]. However, in the JJJ region, most of the cities are located inland and their geographical conditions are not conducive to SR. The results indicated a good fit for spatial spillover of DR in the PRD and JJJ, but not in YRD. Although the PRD and YRD are dominated by light industries, the spillover effects of DR differ. In the YRD, dust pollution in the eastern coastal areas is not serious and the degree of dust pollution in cities that surround megacities is low [46]. The industrial characteristics and geographical location lead to insignificant DR spillover effects.

The ARDL test indicated that the spatial spillover effects of GC and FI on DR and SR, at the city and UA levels, were significant in the long term. Megacities represent the frontier level of technology and economy in China. They provide more scientific and technological support for air pollution control [47], have higher strategic significance, and have greater preferential advantages for policies [48]. The results of the current study indicate pollution reduction spillover effects of megacities on surrounding cities in the three major UAs. Local environmental regulations, climatic and geographical conditions, as well as socioeconomic factors, contribute to the long-term spillover effects of megacities.

There are environmental and policy implications of this study. The results show that infrastructure investment in megacities has a significant positive impact on pollution reduction, and this impact spills over to surrounding cities in the UAs. Since megacities have significant spatial spillover effects on pollution reduction and environmental efficiency of UAs, the overall development of UAs should be promoted, based on green and sustainable development principles [49]. The positive promotional effects of megacities on the environmental quality improvement of UAs can be realized by adjusting the industrial structure of UAs, optimizing infrastructure investment plans, and creating synergistic effects of energy conservation and emission reduction among regions. In addition, policies should guide UAs to take the advantages of megacities, shorten the development gap among cities, and achieve overall regional coordination. The three UAs covered in this study should carry forward their respective development advantages, explore their own distinctive green development modes, and high-quality economic growth paths. Additionally, the government should introduce corresponding policies to achieve coordination between regional economic growth and environmental governance.

This study has some limitations. First, only three major UAs in China were explored, hence the data and spatial spillover effects of other UAs were not collected or analyzed. Moreover, some UAs are excessively large, such as the YRD, which includes more than one megacity and sub-region. Thus, to maintain the comparability of research samples and limitations of data acquisition, only three major UAs were selected as the research objects of this study. Second, due to the differences in cultural backgrounds and geographical environments, the spatial spillover effects of megacities and UAs in China and other countries may show different characteristics. In addition, owing to the limited CCSY data, the current study focused on reductions in SO₂ and dust from industrial pollution. However, a variety of variables (such as NOx, which are also important urban air pollutants) related to infrastructure investment and types of air pollutants should be considered for future research.

The current study makes substantial contributions to the literature. First, an initial effort was made to evaluate the spatial dependence of megacities to explore the spatial spillover effects of pollution reduction in three major UAs. Past studies mainly discussed the spatial spillover effects of air pollution [50]. In contrast, this paper focused on the spillover effects of industrial pollution controls and pollution reduction. The current study also used ANOVA and bootstrap ARDL tests to investigate the short- and long-term pollution reduction spillover effects of megacities on surrounding cities. The bootstrap ARDL model, which is suitable for a small sample size of time series data, successfully explored the effects of spillovers using the integration of infrastructure indicators in three UAs, and also identified long- and short-term causality relationships among pollution reduction and infrastructure investment indicators [51,52]. Finally, the long-term causality among infrastructure investment indicators of megacities and pollution reduction outcomes was also tested. Previous studies have suggested U-shaped [43,44] or N-shaped relationships [45] between economic factors and pollution reduction outcomes. However, the present study revealed a linear relationship between infrastructure investment indicators and pollution reduction outcomes. Overall, the findings of this study indicate that with the increase in infrastructure investment, there is a positive spatial spillover effect of megacities on pollution reduction in UAs over the long term.

Appendix.

Table A1.

Cities in Beijing-Hebei-Tianjin (JJJ), Pearl River Delta (PRD), and Yangtze River Delta (YRD).

JJJ (10)	PRD (9)	YRD (26)
Beijing	Guangzhou	Shanghai	Ningbo
Tianjin	Shenzhen	Anqing	Shaoxing
Baoding	Dongguan	Changzhou	Suzhou_jiangsu
Cangzhou	Foshan	Chizhou	Taizhou_jiangsu
Chengde	Huizhou	Chuzhou	Taizhou_zhejiang
Langfang	Jiangmen	Hangzhou	Tongling
Qinhuangdao	Zhaoqing	Hefei	Wuhu
Shijiazhuang	Zhongshan	Huzhou	Wuxi
Tangshan	Zhuhai	Jiaxing	Xuancheng
Zhangjiakou		Jinhua	Yancheng
		Maanshan	Yangzhou
		Nanjing	Zhenjiang
		Nantong	Zhoushan

Figure A1.

SO2 reduction fitness at the city level.

Figure A2.

SO2 reduction fitness at the urban agglomeration level.

Figure A3.

Dust reduction fitness at the city level.

Figure A4.

Dust reduction fitness at the urban agglomeration level.

Figure A5.

Spatial-time difference of SO₂ reduction.

Figure A6.

Spatial-time difference of dust reduction.