Materials and methods

The cross-regional environmental spillover effect is mainly manifested by the relationship of pollution diffusion, and presents the characteristics of a complex and multi-threaded network. Based on time-series data, the granger causality test can be used to measure cross-regions environmental spillover effect. However, this method only describes the linear “causality” relationship between variables, but it mainly shows a non-linear trend in environmental pollution. To discriminate the non-linear relationship between variables, the non-linear Granger causality test is more suitable (Hiemstra and Jones 1994). Because the stationary sequence is the precondition of the non-linear Granger causality test, it is tested by the Ng-Perron test method (Ng and Perron 2001). If the result is a non-stationary process, then the k-order ($k=1,2,\cdots$) difference sequence will continue to be tested until it is a stationary sequence. Then the BDS test is used to test whether there is a non-linear trend among variables. Finally, a non-linear Granger causality test is performed based on the stationary sequence.

The emphasis on constructing NESE is to determine the relationship between nodes (Scott 2013). Supposing $\left\{X_t\right\}$ and $\left\{Y_t\right\}$ are all strictly stationary process, two VAR models are constructed to test whether there is a non-linear Granger causality for environmental spillover between regions: (1) $X_t={\alpha }_1+\sum _{i=1}^m{\beta }_{1,i}{X}_{t-i}+\sum _{i=1}^n{\gamma }_{1,i}{Y}_{t-i}+{\epsilon }_{1,t}$(1) (2) $Y_t={\alpha }_2+\sum _{i=1}^p{\beta }_{2,i}{Y}_{t-i}+\sum _{i=1}^q{\gamma }_{2,i}{X}_{t-i}+{\epsilon }_{2,t}$(2)

Where ${\alpha }_i,{\beta }_i,{\gamma }_i$ is the estimated parameter, $m$, $n$, $p$, $q$ is the lag order, ${\epsilon }_{it}$$(i=1,2)$ is a residual sequence. Extracting the corresponding information from the residual sequence to analyze the non-Granger causality, if the test result is a non-linear granger causality between$X$and$Y$, then there is a significant spillover effect. Accordingly, a directed link from region$X$to$Y$is made to construct a NESE by analogy.

This paper will describe the network characteristics from both the overall and the individual aspects. Network density and correlation degree are selected to represent the overall characteristics. The former denotes the degree of tightness between nodes, and the greater the value, the stronger the connection between nodes. The latter measures the robustness of the network, and the closer the value is to 1, the stronger the robustness of the network is. Degree centrality, closeness centrality, and betweenness centrality are selected to represent individual characteristics. Degree centrality denotes the number of edges connected with this node, and the larger the value, the more important the role of this node in the network. The closeness centrality measures the degree to which a node is not controlled by other nodes, the larger the value, the stronger the ability of this node to be uncontrolled by other nodes. Betweenness centrality measures the degree to which one node controls other nodes, and the larger the value, the better the node can control the relationship between other nodes.

Environmental pollution may be concentrated in provinces with faster economic growth, and regions with similar locations may have similar variable values. To further measure whether the cross-regional environmental spillover effect (or economic growth) has a significant spatial agglomeration relationship, the global Moran index$I$is introduced to test. Then we have (3) $I=\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}\left(y_i-\bar{y}\right)\left(y_j-\bar{y}\right)/S^2\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}$(3)

where $S^2=\frac{1}{n}\sum _{i=1}^n{\left(y_i-\bar{y}\right)}^2$,$\bar{y}=\frac{1}{n}\sum _{i=1}^n{y}_i$,where$y_i$ represents the level of regional economic growth(or environmental level), $W_{ij}$ represents the spatial weight matrix. If the value of the global Moran index is greater than 0, it indicates that there is a positive spatial correlation between regional economic growth level; on the contrary, it means a negative spatial correlation.

Although the global Moran index can effectively measure the overall level of agglomeration, it cannot describe the spatial distribution characteristics and relations of each region (Wang, Sun, and Wang 2018). That is, it ignores the heterogeneity of local space, and the local Moran’s index can effectively solve this problem, so local Moran’s $I_i$ is used to test whether high and low values tend to agglomerate in the region. Thus, we have (4) $I_i=\left(y_i-\bar{y}\right)\sum _{j=1}^n{W}_{ij}\left(y_j-\bar{y}\right)/S^2.$(4)

If the value of the local Moran index $I_i$ is greater than 0, it indicates that the level of regional economic growth (or environmental level) has positive spatial correlation, and shows spatial agglomeration. On the contrary, spatial correlation is a negative spatial correlation and spatial dispersion is present.

Considering other factors affecting the change of cross-regional environmental spillover effects and economic correlation model (Anselin 1995; Li, Cheng, and Liu 2015; Shi, Yang, and Long 2013; Yu and Jin 2014), this paper attempts to explain the impact from economic and social aspects. On the economic side, besides GDP, the Per Capita GDP (PGDP) should also be taken into account. Once, Jiang et al. (2019) considers the impact of gross domestic product (GDP), PGDP, GDP growth rates and other factors on FDI, because the PGDP of a region is closely related to the development of each industry. This paper uses the logarithmic value of PGDP of each region to reflect the changing trend of this index. In the social aspect, the main considerations: ① WL. Real wage increases may encourage firms to use other factors of production to replace labor costs, thereby enhancing the linkages between industries. WL is expressed by the logarithmic value of the average wage of the provincial employees. ② FDI. The technological and competitive effects of FDI affect the development of regional industries, and then the inter-industry relationship will change. FDI is measured by the proportion of actual foreign investment to GDP of the city. ③ Government Size (GS). GS plays a positive role in industrial development (Jiang 2008), but its non-ideal expansion will also hinder industrial development (Jiang and Li 2013), thus changing the inter-industry correlation. Fiscal indicators are commonly used to measure GS. The proportion of total government expenditure in GDP is used to calculate GS in public finance. Therefore, the proportion of fiscal expenditure to GDP is selected to measure GS. ④ IFL. Improving IFL can significantly reduce transport costs, reduce inter-regional trade costs and improve the allocation efficiency of production factors. It can be predicted that good IFL can enhance the development of industrial linkages, so the per capita road area is used as a measure of IFL. Based on the above factors, the model is constructed as follows: (5) $Y=f(GD{P}_M,PGD{P}_M,W{L}_M,FD{I}_M,G{S}_M,IF{L}_M),$(5)

where $Y$is the binary matrix, which is constructed by non-linear Granger causality test. The explanatory variables $GD{P}_M$,$PGD{P}_M$,$W{L}_M$,$FD{I}_M$,$G{S}_M$,$IF{L}_M$ represents the difference matrix of GDP, PGDP, WL, FDI, GS, and IFL, respectively. Because the explanatory variables are “relational data”, they are similar to each other. Therefore, to avoid measurement errors caused by multiple mutual linear, this paper will use the Quadratic Assignment Procedure (QAP) method of social network analysis to analyze the influencing factors of the NESE.

This paper takes SO₂ as the proxy variable of cross-regional environmental pollution, and chooses the data of 30 provinces (excluding Hong Kong, Macao, Taiwan, and Tibet) from 1992 to 2016 as samples. In 1997, Chongqing was separated from Sichuan and became a municipality city. Therefore, this paper will merge Chongqing and Sichuan, collectively referred to as Sichuan. The data are from China’s provincial “Environmental Situation Bulletin”, “Environmental Statistical Yearbook” and “China Statistical Yearbook 2018”.

Results

Because of the small sample size, this paper uses the more effective Ng-Perron test to test the Horizontal Series (HS) and the First-order Difference Series (FDS) of SO₂ emissions in 30 provinces. The variables of the Ng-Perron test are shown in Table 1. When examining the HS of each province, the results show that the original hypothesis can’t be rejected in HS of Beijing, Tianjin, Hebei, Zhejiang, Fujian, Shandong, Guangdong, Shanxi, Heilongjiang, Jiangxi, Henan, Hubei, Hunan, Guizhou, Yunnan, Tibet, Shannxi, Gansu, Qinghai, Ningxia, Xinjiang, Guangxi and Inner Mongolia, so there is unit root. Therefore, the Ng-Perron test is continued after the first-order difference is applied to the HS. From the results reported in Table 1, all the FDS reject the original hypothesis that they are stationary processes at 5% significance level. Therefore, the next step is the BDS test for the FDS¹. According to the BDS test based on regression residual, the hypothesis of “linear relationship” was rejected at the 10% significant level, that is, there was a significant non-linear relationship between provincial environmental spillovers. In subsequent steps, based on the non-linear Granger causality test (Diks and Panchenko 2006), the environmental spillovers in 30 provinces were examined. Using the residual sequence filtered linearly by the VAR system, the test results of common lag order ($L_x=L_y$) 1–3 are reported². For example, the statistical value of the non-linear Granger causality test between Hebei and Beijing is 1.35* when the time lag is 1 period. That is to say, the original hypothesis that Hebei pollution is not the non-linear granger cause of Beijing pollution is rejected at 10% significant level, that is, Hebei pollution has spillover effects to Beijing. When the time lag is 2 period, the original hypothesis that Shandong pollution is not the non-linear granger cause of Shanghai pollution is rejected at 10% significant level, that is, Shandong pollution has spillover effects to Shanghai. When the time lag is 3 period, the original hypothesis that Shanxi pollution is not the non-linear granger cause of Beijing pollution is rejected at 5% significant level, that is, Shanxi pollution has spillover effects to Beijing. For the remaining values in the non-linear granger causality test, a similar analysis can be made. So far, there is no uniform criterion for the selection of the optimal lag order. If there is at least one test result that significantly rejects the original hypothesis, then the conclusion that there are no spillover effects among variables can be negated. Finally, NESE will build based on these spillover effects. Using the NETDRAW in UCINET software, we build a network model as shown in Figure 1.

Investigating network structure of cross-regional environmental spillover effects and driving factors

All authors

Bin Wang

https://doi.org/10.1080/10962247.2019.1680460

Published online:

05 February 2020

Figure 1. The network of cross-regional environmental spillover effects.

Display full size

The network structure metrics are calculated and the results are shown in Table 2. The network has 126 edges, and its density is 0.136, which shows that the connection of the network is sparser. Due to the implementation of the 12th Five-Year Plan of National Environmental Protection, the effect of pollution reduction is remarkable. Compared with 2010, the total emissions of chemical oxygen demand, ammonia nitrogen, sulfur dioxide, and nitrogen oxides decreased by 10.1%, 9.8%, 12.9%, and 8.6% respectively in 2014. Therefore, the connection of the network decreases accordingly. The correlation degree of the network is 1, which indicates that all provinces are in NESE, and the structure of the network is robust against link failure and varying topology. In terms of degree centrality, the top five provinces are Shanghai, Beijing, Jiangsu, Shandong, and Guangdong. Therefore, Beijing, Shanghai, Guangzhou, Jiangsu, and Shandong have the most significant spatial spillover effects of environmental pollution to other provinces. The top five provinces of closeness centrality: Shanghai, Beijing, Tianjin, Jiangsu, and Zhejiang. It can be seen that these five provinces have the strongest ability not to be controlled by other province’s environmental pollution. The top five of the betweenness centrality are Shanghai, Guangdong, Inner Mongolia, Jiangsu, and Ningxia, which shows that these five provinces mainly play the role of “bridge” and “conduction” in the network. In summary, the high value of degree centrality mainly distributes in the eastern coastal regions. The high values of closeness centrality mainly distribute in the eastern coast and central regions. The high value of betweenness centrality mainly distributes in the northern regions.

Using the software of MATLAB R2014a, the global Moran index values of economic growth and environment can be obtained, as shown in Table 3. All values are positive, and statistically significant at the 5% significance level, so the economic growth and environment have a significant positive spatial correlation. In other words, where spatial distribution is highly concentrated, economic or environmental pollution increases rapidly. To identify the existence of this spatial heterogeneity between the economy and environment, the local Moran index and its agglomeration map are further used to illustrate it.

ArcMap10.0 software is used to calculate the local Moran index values of the economy and environment, and to draw its agglomeration map. The results are shown in Figure 2–3. As shown in Figure 2, the high-value areas of the economy in 2012 and 2016 are mainly concentrated in the eastern coastal regions, including Shanghai, Jiangsu, Shandong, and Guangdong, southern China. These regions have a high level of industrialization or agricultural development, rich scientific research resources or human resources, which can promote economic development, and play spillover effects on the development of the surrounding regional economy. As shown in Figure 3, the high-value areas of the environment in 2012 and 2016 are mainly concentrated in the north-central regions of Inner Mongolia, Liaoning, Hebei, Shanxi, Shandong, Shaanxi, and Jiangsu. These regional industrial activities emit high SO₂, which has a significant environmental spillover effect on the environment of surrounding regions. From 2012 to 2016, Beijing belongs to low and high environmental regions, which indicates that SO₂ emission in Beijing is low in recent years, but environmental pollution in surrounding regions is serious and the environment is poor. Therefore, the Beijing-Tianjin-Hebei region should strengthen the cooperation of environmental pollution prevention and control, and promulgate policy measures of coordinated governance. Hainan is a low-value area with less SO₂ emission and a better environment. In 2012, the high and low regions were mainly concentrated in Guangdong, while in 2016 they were transferred to Guizhou. High and low regions indicate that although environmental pollution in the province is high, it has a limited impact on the surrounding regions. Especially, the “market segmentation” between Guangdong and its surrounding regions is more serious, which, to a certain extent, inhibits the spatial spillover effect of environmental pollution in Guangdong, and produces a strong “siphon effect” on the production factors of economic development in the surrounding provinces.

Investigating network structure of cross-regional environmental spillover effects and driving factors

All authors

Bin Wang

https://doi.org/10.1080/10962247.2019.1680460

Published online:

05 February 2020

Figure 2. Cluster map of economic local Moran’s index.

Display full size

Investigating network structure of cross-regional environmental spillover effects and driving factors

All authors

Bin Wang

https://doi.org/10.1080/10962247.2019.1680460

Published online:

05 February 2020

Figure 3. Cluster map of environmental local Moran’s index.

Display full size

In summary, economic growth and environment show distinct spatial agglomeration characteristics, and the association model varies with regional geographic location. Therefore, to analyze the interaction between two variables related to the relationship, it is feasible to use the network regression model (QAP correlation analysis and QAP regression analysis).

The results of the QAP correlation and regression analysis are shown in Table 4. Based on the difference matrix of GDP, IFL and WL, the correlation coefficients of the matrix of the NESE are 0.737, 0.154 and −0.196, respectively, and that are statistically significant at the 1‰ level. It shows that the more developed the economy of a province is, the more serious the environmental pollution is, and the effect is quite obvious. In areas with good IFL, environmental pollution is not optimistic, but the improvement of WL can reduce environmental pollution. The correlation coefficient between the difference matrix of PGDP and the matrix of the NESE is −0.086, which are statistically significant at the 5% level. It shows that the increase of PGDP can also reduce environmental pollution, but the effect is less obvious than that of the improvement of IFL and WL. The correlation coefficient of the difference matrix of FDI and the matrix of the NESE is −0.049. However, it has not passed the test of significance, so the conclusion that the increase of FDI can improve environmental pollution is still unclear, which is consistent with the previous research (Liu and Lin 2019). The correlation coefficient of the difference matrix of GS and the matrix of the NESE is 0.032. No significant level test has been passed, so the conclusion that the expansion of GS may aggravate environmental pollution is still unclear.

QAP regression analysis can be seen in the fifth and seventh columns in Table 4. From the fifth column, it can be seen that the regression analysis of all variables and the matrix of NESE is not satisfactory. Therefore, this paper further adopts the stepwise regression method, which separates the variables that have not passed the significance test one by one, and finally retains four significant variables. Although the explanatory power of the model decreased from 62% to 61.8%, the QAP regression 2 model is more convincing than the QAP regression 1 and the adjusted R² is 0.617, which shows that the four variables can explain 61.7% of the variation of cross-regional environmental spillover effects. The results are consistent with those of QAP correlation analysis. Among the driving factors, GDP has the strongest explanatory power on environmental pollution, which indicates that economic growth mainly comes at the expense of the environment. Therefore, to reduce environmental pollution, it is necessary to strengthen the implementation of environmental protection policies. Although GDP growth slows down or even decreases, the environment will gradually improve. Compared with the results of QAP correlation analysis, the QAP regression 2 model shows that the coefficient of FDI is significantly negative, and passed the test of the significance level of 1%. It shows that the increase of FDI can improve environmental pollution. Because the knowledge and technology effects of FDI gradually affect the development of regional industries, thereby reducing regional energy consumption and environmental pollution. The explanatory power of WL has also been improved after the variables that have not passed the test of the significance have been precipitated, and passed the test of the significance level of 1‰. It shows that the rise of real wages may promote enterprises to improve their production technology and use other factors of production to replace labor costs, thus improving environmental pollution. The effect of IFL exceeded expectations, but the increase in this value aggravated environmental pollution. This result supports the view that the ecological environment is affected by large-scale infrastructure construction in urbanization (Sun, Wang, and Yao 2015). The reason is that, on the one hand, infrastructure construction may produce a large number of waste-water and waste-gas, resulting in relatively inadequate environmental carrying capacity; on the other hand, infrastructure construction leads to the fragmentation of the original forest vegetation, and land hardening reduces rain infiltration, resulting in deterioration of vegetation growth environment and reduction of vegetation coverage.

Discussion

The NESE is a useful way to understand the environmental spillover among regions; it shows where the environmental spillover goes and where it comes from. Taking 30 provinces in China as an example, this paper uses social network analysis, QAP correlation and regression analysis to explore the association patterns of cross-regional environmental spillover effects and driving factors. Our study finds that the density of the NESE is low, the correlation of environmental spillover effect is relatively small, and SO₂ emissions among regions have sparse spatial correlation and spillover effects. The high value of degree centrality mainly distributes in the eastern coastal regions, which indicates that the correlation of environmental pollution is strong. Because the economy is developed in the eastern coastal regions, material resource consumption is big, and environmental pollution is relatively serious. Therefore, the eastern region is in urgent need of adjusting the industrial structure and reduce pollution both in the region and its surrounding areas (Chen et al. 2019). The high value of closeness centrality mainly distributes in the eastern coast and central regions. This region has the strongest ability not to be controlled by other province’s environmental pollution, and has strong adsorption on resources in its surrounding regions, especially for the major destination of labor mobility, which leads to insufficient environmental carrying capacity and serious pollution in these regions. The high values of betweenness centrality mainly distributes in the northern regions. The regions mainly play the role of “bridge” and “conduction” in the network.

From the analysis of spatial correlation, it can be seen that both environmental pollution and economic growth have a significant spatial agglomeration relationship. In previous studies, Guo et al. (2018), Liu and Lin (2019) argued that there is a significant spatial correlation between environmental pollution of the different provinces. This finding is similar to the results of this study. Where spatial distribution is concentrated, both economic growth and environmental pollution have increased rapidly. The results of the local Moran index show that the high-value areas of regional economies are mainly concentrated in the eastern coastal regions of Shanghai, Jiangsu, and Shandong, and have spillover effects on the development of the surrounding regional economies. Guangdong is also a high-value region of regional economies, and its economic development is quite different from the surrounding regions. The high-value regions of the regional environment include Inner Mongolia, Liaoning, Hebei, Shanxi, Shandong, Shaanxi, and other provinces, and the emission of SO₂ from these regions has an environmental spillover effect on the surrounding areas. The environment of Beijing is a Low-High Outlier, which is at the center of the heavily polluted surrounding regions.

The results of the QAP correlation analysis show that the cross-regional environmental spillover effects are positively correlated with GDP and IFL, but negatively correlated with WL. The results of QAP regression analysis further show that GDP, FDI, WL, and IFL can explain 61.7% of the variation of cross-regional environmental spillover effects. Among them, the regression coefficient of GDP is positive (0.767), which has the strongest explanatory power for environmental pollution, followed by the regression coefficient of IFL is positive, but the regression coefficient of FDI and WL is negative.

Conclusion and policy implications

By using the non-linear granger causality test, this paper verifies that there was a significant non-linear relationship between provincial environmental spillovers, then the NESE will build based on these spillover effects. Furthermore, an empirical analysis of the factors affecting the structural variation of NESE has been taken by QAP correlation and regression analysis. Through analyzing the NESE, the results show that there is an environmental spillover effect among regions. Second, the paper studies the centrality of the NESE and find that degree centrality mainly distribute in the eastern coastal regions at the regional level. The results of the Moran index shows that economic growth and environmental pollution have spatial correlation effects. The regression analysis finds that the regression coefficient of GDP and IFL is positive, but the regression coefficient of FDI and WL is negative.

On this basis, some specific policy recommendation is proposed. First, to reduce pollution emissions, the government needs to coordinate with the surrounding areas to control the environment to minimize the spillover effects of the surrounding regions on their environmental pollution. The mode of production should be further improved to promote the balanced development of cross-regional economies. To achieve the overall goal of medium-to-high growth and sustainable improvement of the ecological environment, we should optimize the network of the cross-regional environmental spillover effect to boost the reasonable development of the economy.

Second, the restraint of the carrying capacity of the regional ecological environment should be fully considered. Due to the limited carrying capacity of the regional environment, excessive pursuit of economic growth will aggravate environmental pollution. To promote the development of regional economies in southeastern, we must implement measures of innovation-driven and cooperative governance of the environment under the condition of increasingly intensified environmental and resource constraints. The eastern coastal regions should continue to play the role of economic spillover effects to promote the development of surrounding regions. At the same time, we should pay more attention to the problem of environment pollution.

Finally, environmental benefits should also be emphasized in infrastructure construction. An environmental protection policy should be strictly enforced to minimize waste discharge in the process of infrastructure construction. Infrastructure construction should try to maintain the original ecological environment, reduce the ground hardening area, increase the water seepage capacity, without affecting the original rivers and lakes, and enhance the water storage capacity of urban. We should evacuate the population density of big cities, improve the infrastructure of surrounding regions, and alleviate the pressure of insufficient infrastructure in big cities.