Urbanisation and the geographic concentration of industrial SO 2 emissions in China

Abstract

Chinese

This paper studies the mechanisms linking urbanisation and industrial SO₂ emissions using panel data that enable us to trace the environmental impact of the urban transformation of China’s economy from a rural to a largely urban society. We provide evidence of stylised facts on spatial patterns and temporal changes of industrial SO₂ emissions in China. Over time, industrial SO₂ emissions show increasing levels in small and medium-sized cities but slightly decreasing levels in some large cities. The results show that urbanisation is one of the main driving forces behind emissions and the increase in the urbanisation level is likely to exacerbate emissions. Emissions are more sensitive to industrialisation than urbanisation, indicating that industrialisation remains a key industrial SO₂ pollution contributor in China. Industrial emissions abatement policies in China should be designed by considering the spatial differences of emissions and the pace of urbanisation, although the increase of urbanisation is associated with benefits such as poverty reduction and economic growth. The reduction of industrial SO₂ emissions requires coordinated approaches by adjusting the pace of urbanisation.

Keywords

Chinese cities industrialisation urbanisation

Introduction

China’s rapid economic growth has come with increasing concerns regarding air pollution, including the emission of sulphur dioxide (SO₂). SO₂ is a colourless highly reactive gas which negatively effects human health, the environment and the economy (Currie and Neidell, 2005; Ebenstein, 2012). SO₂ has adverse effects on the respiratory system, infant mortality, birth defects and even worker productivity (Brajer and Mead, 2004; Chen et al., 2013). Lu et al. (2010) find that the trend of SO₂ emissions correlates with the trend of acid rain pH in China. Acid rain, caused mainly by SO₂ emissions, costs 30 billion yuan in crop damage, which amounts to 1.8% of the value of agricultural output (World Bank, 2007). SO₂ is also a major source of particulate matters with diameter less than 2.5 μm (PM_2.5) and is a radioactive forcing agent that affects climate change (Masiol et al., 2014).

The largest sources of SO₂ emissions are fossil fuels such as coal, oil and natural gas combustion in power plants (73%) and other industrial facilities (20%) (Environmental Protection Agency, 2014). In 2000, China emitted 20.4 teragrams of SO₂ (1 teragram = 1 million tonnes), 45% of total Asian SO₂ emissions (Streets et al., 2003). China’s total SO₂ emissions had increased by 53% from 2000 to 2006 (Lu et al., 2010). As a result of the country’s commitment to SO₂ reduction, the trend changed in 2009 with SO₂ emissions reduced by 13.1% from the level in 2005 (Xu, 2011; Zhang et al., 2015).

One potentially significant component that may contribute to SO₂ emissions is urbanisation. As a process of increasing the non-agricultural population in cities, urbanisation is manifested not only by population migration from rural to urban areas but also by the shift of the labour force from the agricultural sector to non-agricultural sectors (Poumanyvong and Kaneko, 2010). China has experienced a steady rise in its urbanisation rate from 40.53% in 2003 to 51.27% in 2011. More importantly, the National New Urbanisation Plan (2014–2020) initiated in 2014, one of China’s most contentious projects (World Bank and the Development Research Center of the State Council, China (DRC), 2014), has set an ambitious goal: to increase the urbanisation rate to 60% by 2020 (or more than 100 million people moving from China’s rural areas to cities) to promote economic growth.

Urbanisation has served as an essential engine for China’s economic development (World Bank and DRC, 2014). Major city clusters formed by urbanisation lead to an agglomeration economy through the concentration of consumption, economic activity and labour forces (Liu et al., 2015). However, as one of the major emerging market countries, China has paid an enormous price for environmental pollution. The cost of China’s environmental degradation and resource depletion is estimated at approximately 9% of its GDP (World Resources Institute (WRI), 2014). The conflict between the economic benefits and environmental costs of urbanisation raises the question of the degree to which urbanisation drives up environmental emissions. Deeper reasons behind air pollution emissions have not been fully explored (Astaraie-Imani et al., 2012; Poumanyvong and Kaneko, 2010). Some studies reveal that urbanisation contributes to the rise in pollutant discharges (Cherniwchan, 2012; Zhang et al., 2014). Some researchers argue, however, that urbanisation has little impact on environmental emissions. Sadorsky (2014) shows that urbanisation may increase or decrease carbon dioxide (CO₂) emissions.

Furthermore, emissions patterns and factors behind industrial SO₂ production have not been thoroughly investigated with consistent results. Only a few studies on SO₂ emissions can be found, namely the works of Antweiler et al. (2001), Cole and Neumayer (2004), Frankel and Rose (2005), Cherniwchan (2012) and Zhang et al. (2014). Cole and Neumayer (2004) argue that urbanisation is not a significant determinant for SO₂ emissions. It remains unclear what is the most significant contributor to emission production and whether the rapid urbanisation rate, high energy consumption, or high ratio of manufacturing share significantly contribute to the substantial rise in industrial SO₂ emissions. It is also unknown whether there is a turning point for emissions when economic growth reaches certain threshold values in a city. It is therefore worthwhile to further investigate the relationship between urbanisation and industrial SO₂ emissions and provide useful guidance regarding China’s urbanisation policy.

Accompanied by urbanisation, new urban residents seek employment and interact with the labour demand mainly in the secondary and service sectors in a city, which may accelerate industrialisation. Industrialisation reflects a shift from agricultural production to industrial production and services. In addition, population concentration in cities also plays an important role in driving growth in energy consumption. The energy needed for manufacture is likely to lead to a large portion of the growth in the demand for energy when a city expands (Poumanyvong and Kaneko, 2010). The increase in energy consumption will have a significant impact on environmental emissions. Zhang and Cheng (2009) reveal that there is a unidirectional causality from energy consumption to carbon emissions. However, a decrease in energy intensity can contribute to a decrease in carbon emissions (Dhakal, 2009). The net impact of energy consumption on emissions remains unknown.

Existing studies on emissions have largely been conducted at the aggregate level, such as the country or state level, which generally have to deal with issues such as extreme values or outliers across observations. For example, the sizes of provinces in terms of population or GDP vary substantially across China, from small provinces such as Hainan to large ones such as Zhejiang. Such large variations are likely to lead to potential heteroskedasticity and biased estimates. Analysis at the city level in this study can avoid such heterogeneity problems.

To avoid estimate bias, we include the Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model¹ in our analysis. The model has identified three key driving forces of environmental pollution: population, affluence and technology. Large cities usually have more pollutants discharged. Owing to agglomeration economies, however, large cities may have higher energy efficiency which results in lower per capita emissions than small and medium-sized cities. Affluence is highly related to urbanisation and environmental emissions (Henderson and Wang, 2007; Liu et al., 2015; Zhang, 2002; Zhang and Cheng, 2009).

Furthermore, we test the hypothesis of the inverted U-shaped relationship or Environmental Kuznets Curve (EKC) between SO₂ emissions and affluence. The environmental EKC hypothesises that environmental emissions increase in the early stages of economic growth and the trend reverses when income per capita reaches some high level (Stern, 2004). However, some researchers argue that the turning point depends on the type of pollutants or emissions. For instance, Gassebner et al. (2011) find evidence that supports the EKC hypothesis for water pollution, but not for air pollution, across 120 countries. Auffhammer and Carson (2008) reject the static EKC evidence based on Chinese province-level data. By including the income quadratic effect this research explores if turning points exist in SO₂ emissions. In addition, emissions may vary by city size. To control for the urban hierarchy and heterogeneity issue, city population and its quadratic effect are included in the study.

Finally, to consider the characteristics of the transition economy, we include foreign investment. He et al. (2012) discuss the impact of economic transition factors such as globalisation and marketisation and argue that their impact is only statistically significant in the coastal and central cities. The environmental degradation effect by marketisation was cancelled by the effect of globalisation.

In sum, based on a city data set at or above the prefectural level,² this research explores the impact of urbanisation on SO₂ emissions in China. A concise study on urbanisation and other driving forces on SO₂ emissions would enrich the current literature and provide important environmental policy implications for the environmental challenges brought by the new wave of urbanisation. Policy priority can be given to handling the most significant emissions driving factors or production activities.

Methods

A minimalist model

This study uses a simple model to test the components that influence SO₂ emissions (E) at the city level. Specifically, we test three driving forces for pollutant emissions – urbanisation (U), energy consumption (C) and industrialisation (S) – while controlling other factors X, such as population, economic growth and technological progress identified in the STIRPAT model. Two additional control variables are included in the study, one accounts for the industrial reshuffling effect and one for foreign investment. In addition, the BP Energy Outlook explains the factors behind energy consumption, asserting that population growth is one of the key drivers of energy demand and thus leads to the increase in carbon emissions in the long run (BP, 2013). The variable for population is also used to control for the size effect in the regression. The scale effects of both affluence and population growth may act to increase carbon emissions given no change in other factors such as technology progress. The increase in population density can reduce per capita emissions because of economies of scale. Furthermore, factors such as geographic features and China’s transition economy are also considered in the model. We, therefore, separate those factors from urbanisation by including factors on industrial structural changes and electricity consumption as well as other factors so that our models can catch the effects of urbanisation on the emissions. Following the practice of Dietz and Rosa (1997), we apply a Cobb-Douglas function form in the expanded emissions equation:

E = α U^{γ} C^{φ} S^{\emptyset} X^{β}

(1)

We start with the conjecture that pollutant emissions are mainly determined by the urbanisation rate, energy consumption and industrialisation while controlling other potential factors. We rewrite equation (1) in a general empirical econometric form. The SO₂ emissions in city i in year t is as follows:

e_{i, t} = α + γ u_{i, t} + φ c_{i, t} + ϕ s_{i, t} + X_{i, t}^{'} β + μ_{t} + η_{i} + ε_{i, t}

(2)

In equation (2), i is the individual city ( $i = 1, \dots, N)$ ; t represents years in the study period ( $t = 1, \dots, T)$ ; $e_{i, t}$ represents the natural logarithm of SO₂ emissions in city i at time t; $α$ is the constant; $u_{i, t}$ is the urbanisation measured by the ratio of non-agricultural population; $c_{i, t}$ is the energy consumption and $s_{i, t}$ is the industrialisation; $X_{i, t}$ is a vector of control variables; $μ_{t}$ denotes the time effect controlling for general shocks from macroeconomic or technological issues; $η_{i}$ is the fixed effect that controls for the unobserved time-invariant city-specific characteristics; and $ε_{i, t}$ is the idiosyncratic error term. $ε_{i, t}$ is assumed to be independent and identically distributed (i.i.d.) random variables with zero means and constant variances. The fixed effect estimation of panel data model can avoid estimation bias by excluding the impact of factors without variety overtime.

In addition to the static model, for robustness checks this study also employs a dynamic panel data model to investigate the impact of the lagged pollution variable on SO₂ emissions. Following equation (2), we obtain the dynamic model:

e_{i, t} = α + k e_{i, t - 1} + γ u_{i, t} + φ c_{i, t} + ϕ s_{i, t} + X_{i, t}^{'} β + μ_{t} + η_{i} + ε_{i, t}

(3)

where $e_{i, t - 1}$ is the lagged pollutant emissions variable used to account for the persistence of emissions over time. The equation is estimated through Generalised Method of Moments (GMM) estimators using instrument variables to solve the omitted variable problem and measure error issues.

Furthermore, this paper attempts to test the EKC hypothesis of income and the quadratic effect of population. The reasons for the turning point are likely to be the scale effect of population and technological progress with the accumulation of human capital. Recent work by Henderson and Wang (2007) shows that technology improvements help bigger cities to accumulate human capital relative to smaller ones based on worldwide data for all metropolitan areas with a population over 100,000 from 1960 to 2000. A larger environmentally aware population in large cities may also exert more political pressure to achieve emissions reductions (Damania et al., 2003). We extend equation (2) by including the quadratic effect of both population and income:

e_{i, t} = α + γ u_{i, t} + φ c_{i, t} + ϕ s_{i, t} + X_{i, t}^{'} β + θ p_{i, t}^{2} + ϑ w_{i, t}^{2} + μ_{t} + η_{i} + ε_{i, t}

(4)

where $p_{i, t}^{2}$ is used to measure the quadratic effect of the total city population and $w_{i, t}^{2}$ is added to account for the quadratic effect of income represented by GDP per capita.

Data and measurement

The city level panel data used in this research are compiled from various official sources, including China City Statistical Yearbook (CCSY), China Population & Employment Statistics Yearbook (CPESY), and Chinese Economy Network (CEN). This research constructs a panel for 286 prefectural-level cities by matching data sets over the same period by city name from 2003 to 2011. Sixteen cities are excluded because of a lack of data for the study years.

The dependent variable, SO₂ emissions, is measured by industrial SO₂ emissions (SO₂). Industrial SO₂ emissions refer to the SO₂ emitted from industrial activities such as the production of manufacturing goods. Emissions do not include SO₂ emissions from motor vehicles and households. For independent variables, urbanisation is measured as the ratio of non-agricultural population to total population in a city (Urbansh) by considering the official definition of urbanisation and the approach by Sadorsky (2014). Electricity consumption (Enercon) is used as a proxy for energy consumption. Approximately 70% of electricity generation is dominated by the burning of coal, which is one of the major sources of SO₂ emissions. Industrialisation is measured by the share of the secondary industry sector (Indush).

For control variables $X_{i, t}$ , population is denoted by city population (Pop) and the real GDP per capita (Pgdp) is a proxy for economic growth. Technology progress is denoted as human capital accumulation by the number of university faculty (Univ) and number of employees in the science and technology industry (Tech). GDP share of the service sector (Svcsh) is used to account for the industrial reshuffling effect on emissions and the number of foreign investment contracts (Ficon) is used to control the effect of foreign investment on pollution for instance the pollution heaven hypothesis, which hypothesises that pollution may be transferred through foreign direct investment by relocating pollution plants from developed economies to developing economies. City built area (Area) is included to control the boundary change effect. The dollar value variables including GDP and GDP per capita are deflated by GDP deflators at the 2005 constant price (2005 = 100).

Results

The distribution of SO₂ emissions

Two kernel density estimates of the log SO₂ emissions indicate a distribution that is close to Gaussian with a slightly right skew, as shown in Figure 1. The similar curve shapes suggest that the distribution of emissions across cities has no systematic structural break over the two periods. The upper tail represents the portion of cities with extremely high SO₂ emissions, while the lower tail represents cities with exceptionally low SO₂ emissions.

Figure 1.

Kernel density estimates for log SO₂ emissions in 2003 (dashed line) and 2011 (solid line).

Figure 2 shows the spatial distribution of SO₂ emissions in 286 Chinese cities in 2011. The size of the circles represents the emissions amounts in each city, as shown in the legend of the map. This map indicates that there is an unbalanced spatial distribution of SO₂ emissions across cities. Emissions are mainly clustered in coal-rich northern China, the more developed eastern region, the Pearl River delta in the south, and some parts of the southwest region. Figure 3 illustrates the emissions changes. Overall SO₂ emissions increased by 22.6% during the sample period; however, the variation among cities is noticeable.

Figure 2.

The spatial distribution of SO₂ emissions in key Chinese cities in 2011.

Figure 3.

The spatial distribution of SO₂ emissions changes in key cities from 2003 to 2011.

Statistical results

Table 1 presents summary statistics for the variables. The mean (median) of the dependent variable SO₂ emissions (SO₂) in the sample is 66,493 (53,177) tonnes. The standard deviation of the variable is smaller than its mean. The independent variable urbanisation (Urbansh) has a mean (median) of 61 (63%) with a small standard deviation.

Table 1.

Descriptive statistics.

Definition	Variables	N	Mean	Median	Std dev.	Min	Max	Data sources
Dependent variables
SO₂ emissions (tonnes)	SO₂	2574	66,493	53,177	63,662	92	683,162	CCSY
Independent variables
Share of non-agricultural population (%)	Urbansh	2574	61.31	63.06	25.04	12.17	100	CPESY & CEN
Electricity consumption (10,000 kwh)	Enercon	2574	629,676	294,131	1,121,837	8055	13,396,200	CEN
Share of industrial output (%)	Indush	2574	51.76	52.60	12.56	8.05	90.97	CEN
Share of service output (%)	Svcsh	2574	41.19	40.29	10.64	8.58	80.89	CEN
Total city population (10,000 persons)	Pop	2574	129.66	85.99	161.03	14.08	1770.60	CPESY
GDP per capita (RMB yuan)	Pgdp	2574	26,639.71	21,970.04	18,848.13	1715.42	206,268.60	CCSY
Number of university faculty (persons)	Univ	2574	3880.45	1372	7697.60	0	58,735	CEN
Number of people in science & technology industry (10,000 persons)	Tech	2574	0.73	0.18	2.68	0.01	50.32	CEN
Foreign investment contract number (count)	Ficon	2574	87.64	11	334.04	0	4329	CEN
City built area (square km)	Area	2574	98.31	57	149.91	5	3371	CCSY

Notes: CCSY, China City Statistical Yearbook; CPESY, China Population & Employment Statistics Yearbook; CEN, Chinese Economy Network.

Table 2 reports the effects of urbanisation (logUrbansh) on SO₂ emissions while controlling other factors such as energy consumption, industrialisation, demographic, economic, technological and opening condition factors. The time effect is used to control for possible shocks occurring in some years, for example, the financial crisis of 2008 and the widespread installation of SO₂ scrubbers over the period from 2006 to 2008. Table 2 shows that the signs and magnitudes of urbanisation (logUrbansh) in all columns are remarkably consistent and statistically significant at the 5% level or lower. The estimated results on urbanisation indicate that the acceleration of urbanisation in Chinese cities significantly contributed to the rise in SO₂ emissions. In the baseline model, the estimate coefficient of urbanisation is 0.101 with a standard error of 0.042, statistically significant at the 1% level; this indicates that an increase of one percentage point in the urbanisation rate is likely to lead to a 0.1% rise in SO₂ emissions while industrialisation (logIndush) has stronger magnitudes than urbanisation.

Table 2.

Fixed effects estimation of urbanisation on SO₂ emissions.

Variables	Dependent variable = log(SO₂ emissions)
Variables	(1)	(2)	(3)	(4)	(5)	(6)
logUrbansh	0.101**	0.116***	0.087**	0.102**	0.102**	0.148***
	(0.042)	(0.043)	(0.042)	(0.043)	(0.043)	(0.043)
logEnercon	0.029*	0.026	0.030*	0.027	0.027	0.038**
	(0.017)	(0.017)	(0.017)	(0.017)	(0.017)	(0.016)
logIndush	0.165***	0.164***	0.127**	0.126**	0.126**	0.097
	(0.063)	(0.063)	(0.064)	(0.063)	(0.064)	(0.064)
logSvcsh	−0.011	−0.010	−0.021	−0.020	−0.020	−0.029
	(0.044)	(0.044)	(0.044)	(0.044)	(0.044)	(0.044)
logPop	0.001	0.514**	0.005	0.526**	0.526**	0.621***
	(0.039)	(0.223)	(0.039)	(0.222)	(0.230)	(0.229)
logPgdp	0.056***	0.056***	1.014***	1.019***	1.019***	1.171***
	(0.021)	(0.021)	(0.234)	(0.234)	(0.234)	(0.232)
logUniv	−0.036**	−0.039**	−0.036**	−0.039**	−0.039**	−0.033*
	(0.018)	(0.018)	(0.018)	(0.018)	(0.018)	(0.018)
logTech	−0.059***	−0.059***	−0.054**	−0.054**	−0.054**	−0.050**
	(0.022)	(0.022)	(0.022)	(0.022)	(0.022)	(0.022)
logFicon	0.003	0.003	0.002	0.002	0.002	0.007*
	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)	(0.004)
logPopsq		−0.050**		−0.051**	−0.051**	−0.056**
		(0.022)		(0.021)	(0.022)	(0.022)
logPgdpsq			−0.049***	−0.049***	−0.049***	−0.057***
			(0.012)	(0.012)	(0.012)	(0.012)
logArea					0.000	0.008
					(0.023)	(0.023)
Constant	9.394***	8.137***	4.928***	3.630***	3.630***	2.290*
	(0.544)	(0.764)	(1.213)	(1.329)	(1.329)	(1.265)
Time effects	Yes	Yes	Yes	Yes	Yes	No
Observations	2574	2574	2574	2574	2574	2574
Number of cities	286	286	286	286	286	286
F-value	170.7	170.6	171.8	171.7	170.7	167.1
Log likelihood	846.2	849.4	855.8	859.0	859.0	814.0
						Model 8

Notes: Columns (1)–(4) and (6) report two-way fixed effects of panel data regressions, with standard errors in parentheses. Column 1 is the baseline model, and columns (2)–(4) expand the baseline model by adding quadratic terms of population (logPopsq), income (logPgdpsq), and both. Column (5) includes the variable on city built areas to control for the potential estimation bias caused by the change of city boundary and column (6) does not include time effects. *, **, and *** indicate two-tailed statistical significance at 10, 5, and 1% levels, respectively.

Column (2) in Table 2 extends the baseline model by capturing the quadratic effect of city population on SO₂ emissions while column (3) explores the quadratic effect of income measured in GDP per capita, which is to test the environmental EKC hypothesis. The impact of income on SO₂ emissions reaches a turning point when GDP per capita exceeds 30,512 renminbi (RMB) yuan.³ Column 4 investigates the quadratic effect of both income and population on SO₂ emissions. The turning point for income is 30,782 RMB yuan,⁴ respectively. Taken together, these results present a provocative picture of the driving forces of SO₂ emissions. Urbanisation is one of the important factors for SO₂ emissions. Column (1) of Table 2 shows that there are three independent variables at the 1% significance level: industrialisation, the affluence component and technological progress. There are two independent variables at the 5% significant level: urbanisation and number of university faculty. We thus can conclude that urbanisation contributes to the SO₂ emissions, although it affects emissions at a less significant level than industrialisation and economic growth.

Finally, urban boundary change may lead to the variation of urbanisation as some rural population near city boundaries may automatically become urban residents with urban expansion. To deal with boundary change issue, we add a variable on city built area in columns (5)–(6). The results show that such factor is not statistically significant and negligible in the model. The results of fixed effect estimation without time effect in column (6) are consistent with the results of baseline model. The F-test results imply that two-way fixed effect regression models should be adopted.

Robustness checks, discussion and policy implications

Robustness checks

To check whether our results are sensitive to model specifications and estimation approaches, we conduct a series of robustness checks. To consider the geographical characteristics, we separately investigate the different impact of urbanisation across three different regions, i.e. the western region, the middle region and the eastern region.⁵ The results in Table 3 indicate that urbanisation is statistically significant at the 1% level for the western region while its impact is not statistically significant in middle and eastern regions. In other words, urbanisation in the western region has the largest impact on emissions while it demonstrates little impacts on other two regions. One potential reason behind such result is that development of urbanisation in the western region with median of 43.7% is significantly lagged behind the one in eastern region with median of 76.7%.

Table 3.

Fixed effects estimation of the urbanisation effect on SO₂ emissions in three regions

Variables	Dependent variable = log(SO₂ emissions)
	West	Middle	East	West	Middle	East
	(1)	(2)	(3)	(4)	(5)	(6)
logUrbansh	0.274***	0.050	0.114	0.254***	0.047	0.100
	(0.082)	(0.068)	(0.073)	(0.082)	(0.068)	(0.074)
logEnercon	0.047**	0.047*	−0.022	0.058**	0.046*	−0.028
	(0.024)	(0.025)	(0.041)	(0.024)	(0.025)	(0.041)
logIndush	0.090	−0.094	0.623***	0.040	−0.099	0.594***
	(0.112)	(0.083)	(0.150)	(0.112)	(0.083)	(0.151)
logSvcsh	0.154**	−0.062	−0.022	0.110	−0.062	−0.041
	(0.078)	(0.059)	(0.101)	(0.079)	(0.059)	(0.102)
logPop	0.017	0.114*	0.087	0.020	0.123**	0.083
	(0.070)	(0.059)	(0.068)	(0.069)	(0.059)	(0.068)
logPgdp	0.123***	−0.006	0.033	1.433***	0.590*	1.061
	(0.036)	(0.024)	(0.040)	(0.422)	(0.349)	(0.662)
logUniv	−0.008	0.012	−0.091***	−0.005	0.007	−0.088***
	(0.030)	(0.029)	(0.033)	(0.030)	(0.029)	(0.033)
logTech	−0.064	−0.041	−0.051	−0.062	−0.037	−0.039
	(0.049)	(0.027)	(0.041)	(0.049)	(0.027)	(0.042)
logFicon	0.010*	0.011*	−0.026**	0.008	0.011*	−0.027**
	(0.006)	(0.006)	(0.012)	(0.006)	(0.006)	(0.012)
logPgdpsq				−0.07***	−0.030*	−0.050
				(0.022)	(0.018)	(0.032)
Constant	7.250***	10.132***	8.876***	1.378	7.216***	3.873
	(0.772)	(0.662)	(0.936)	(2.036)	(1.830)	(3.350)
Observations	765	900	909	765	900	909
Number of cities	85	100	101	85	100	101
F value	183.1	132.2	114.9	185.0	132.1	114.8
Log likelihood	159.0	460.3	261.2	164.5	461.9	262.5
						Model 8

Notes: Columns (1)–(3) report two-way fixed effects of west, middle and eastern regions, with standard errors in parentheses. Columns (4)–(6) add quadratic terms of income (logPgdpsq). *, **, and *** indicate two-tailed statistical significance at 10, 5, and 1% levels, respectively.

Table 4 reports regression results that check the sensitivity to changes in the base specifications. The regression in column (1) of Table 4 drops the variables that control for energy consumption and industrial structure. The result is an increase in the estimated effect of urbanisation, which remains significant at the 5% level with no appreciable change in the coefficient magnitude. The regression result in column (2) shows that energy consumption becomes statistically significant at the 10% level without a substantial change in the magnitude, indicating that energy consumption is not sensitive to the change in functional form. The findings for industrialisation in columns (3)–(4) are largely unchanged.

Table 4.

Fixed effects estimation of panel data for the impact of urbanisation, energy consumption and industrialisation on SO₂ emissions.

Variables	Dependent variable = log(SO₂ emissions)
Variables	(1)	(2)	(3)	(4)
logUrbansh	0.099**
	(0.043)
logEnercon		0.029*
		(0.017)
logIndush			0.136**	0.143***
			(0.063)	(0.053)
logSvcsh			−0.009
			(0.044)
logPop	0.572***	0.442**	0.461**	0.463**
	(0.221)	(0.218)	(0.218)	(0.217)
logPgdp	1.114***	1.158***	1.064***	1.061***
	(0.231)	(0.230)	(0.233)	(0.233)
logUniv	−0.0377**	−0.041**	−0.038**	−0.039**
	(0.018)	(0.018)	(0.018)	(0.018)
logTech	−0.052**	−0.050**	−0.049**	−0.049**
	(0.022)	(0.022)	(0.022)	(0.022)
logFicon	0.002	0.001	0.002	0.002
	(0.004)	(0.004)	(0.004)	(0.004)
logPopsq	−0.055**	−0.049**	−0.046**	−0.046**
	(0.021)	(0.021)	(0.021)	(0.021)
logGdpsq	−0.053***	−0.056***	−0.051***	−0.051***
	(0.012)	(0.012)	(0.012)	(0.012)
Constant	3.699***	3.907***	4.172***	4.118***
	(1.302)	(1.297)	(1.317)	(1.290)
Time effects	Yes	Yes	Yes	Yes
Observations	2574	2574	2574	2574
Number of cities	286	286	286	286
F-value	207.3	187.2	182.8	189.2
Log likelihood	853.2	851.8	854.4	854.3

Notes: Standard errors are in parentheses; *, **, and *** indicate two-tailed statistical significance at 10, 5, and 1% levels, respectively.

This study also checks the model sensitivity with an alternative estimation approach. Table 5 presents the results using robust dynamic system GMM estimators. The two-period lagged levels of the regressors are applied as instruments for the regressions. Consistent with the results in Table 2, the estimate results show that industrialisation is statistically significant with a positive relationship with emissions. Urbanisation and energy consumption also have positive signs as expected, but become statistically insignificant with robustness. The lower panel of Table 5 includes post estimation tests for autocorrelation and the validity of instruments. AR (2) is the Arellano-Bond tests for the second order autocorrelation in the first differenced errors. All p-values of the AR (2) test in the four specifications are larger than 0.1, indicating that there is no evidence of autocorrelation at conventional GMM moment conditions. The Sargan test or Hansen J test is the test used for over identifying restrictions; it is a chi-square test to judge whether the residuals are correlated with the instruments. A rejection of the null hypothesis from the over test implies that the model or instruments may be misspecified. The Sargan test in column (1) shows that there is no evidence of over-identification issues at the 5% level. But the null hypothesis is rejected at the 10% level, indicating that the one-step GMM model is not appropriate for the 10% level in column (1) of Table 5. We further consider three specifications and the Hansen J tests show no evidence of model misspecification problems at any conventional significance levels. These post estimation results indicate that the dynamic panel models in Table 5 are correctly specified with valid GMM estimators. In summary, the results regarding the independent impacts of urbanisation on SO₂ emissions are robust while controlling for traditional determinants of emissions.

Table 5.

Dynamic GMM estimation results.

Variables	Dependent variable = log(SO₂ emissions)
Variables	(1)	(2)	(3)	(4)
logSO2_t-1	0.936***	0.947***	0.947***	0.946***
	(0.037)	(0.019)	(0.019)	(0.018)
logUrbansh	0.0522	0.0357	0.0365	0.0333
	(0.047)	(0.049)	(0.050)	(0.049)
logEnercon	0.052	0.009	0.008	0.009
	(0.061)	(0.014)	(0.014)	(0.014)
logIndush	0.145	0.085**	0.077*	0.089**
	(0.103)	(0.043)	(0.043)	(0.044)
logSvcsh	−0.033	0.0737	0.075	0.080
	(0.111)	(0.051)	(0.050)	(0.051)
logPop	0.005	0.021	0.026	0.039
	(0.085)	(0.027)	(0.028)	(0.064)
logPgdp	−0.187*	−0.010	0.231	−0.008
	(0.106)	(0.017)	(0.327)	(0.017)
logUniv	0.007	−0.004	−0.006	−0.003
	(0.025)	(0.012)	(0.012)	(0.012)
logTech	0.063	−0.001	−0.0003	−0.001
	(0.045)	(0.014)	(0.0141)	(0.014)
logFicon	−0.025	−0.009	−0.010	−0.010
	(0.022)	(0.007)	(0.008)	(0.007)
logPgdpsq			−0.012
			(0.016)
logPopsq				−0.002
				(0.006)
Constant	0.954	−0.116	−1.304	−0.211
	(0.903)	(0.374)	(1.679)	(0.373)
Time effects	Yes	Yes	Yes	Yes
Observations	2002	2002	2002	2002
Number of cities	286	286	286	286
Sargan test (p-value)	0.062
AR2 test (p-value)	0.111	0.335	0.331	0.334
Hansen J test (p-value)		0.214	0.252	0.226

Notes: Robust standard errors are in parentheses. *, **, and *** indicate two-tailed statistical significance at 10, 5, and 1% levels, respectively. AR (2) test, Sargan test and Hansen J test for over-identification are shown at the bottom of the table. All variables are used in natural logarithm form.

Discussion

The positive urbanisation elasticity of the emissions in the baseline model shows that urbanisation increases SO₂ emissions. Our results are similar with the finding that urbanisation leads to higher pollutant emissions such as CO₂ emissions in middle-income group countries such as China (Poumanyvong and Kaneko, 2010). It is likely that urbanisation accompanied by the increase in population within cities and the shifting of employment from the agricultural sector to the industrial and service sectors has contributed to the increase in SO₂ emissions because of the increase in manufacturing production activity and energy consumption. As a transition from agricultural production to industrial production or a shift toward possible pollution intensive industrial production, industrialisation is a central determinant of changes in environmental quality based on evidence from sulphur emission data for 157 countries over the period 1970–2000 (Cherniwchan, 2012). This phenomenon illustrates that industrial activity-related emissions are dominant in cities in developing countries such as China. China has surpassed the USA to become home to the world’s largest manufacturing sector measured by its value-added. Although the manufacturing’s output share of GDP has remained stable over a decade, there has been a large increase in the absolute level of manufacturing employment. The industrialisation elasticity of emissions is larger than urbanisation and energy consumption. The results regarding industrialisation suggest that the increase in the share of secondary industry is highly associated with a substantial increase in the level of SO₂ emissions. The positive link between industrialisation and SO₂ in our study is consistent with the study by Cherniwchan (2012).

The service share factor is intended to measure the effect of composition on SO₂ emissions. The share of service sector value-added (logSvcsh) has a negative sign in the baseline model (column 1) in Table 2, which suggests that the service share factor may reduce environmental emissions but that this effect is not statistically significant. The estimate result is different from He and Wang (2012), who claim that upgrading the industrial structure can significantly reduce environmental emissions. Instead, similar with CO₂, our result is consistent with research by Martinez and Silveira (2013), who argue that the role of industrial structural changes in CO₂ emissions reduction is minor based on Swedish manufacturing industries. A recent study by Qi et al. (2014) also implies that the policy on sector shifting from industry to the service sector has only a limited impact on China’s trade-embodied CO₂ emissions without a decrease in the trade surplus. Our result suggests a limited impact of industrial structure change on SO₂ emissions.

As one important factor for emission reduction, the technique effect measured through the number of university faculty and number of people in the science and technology industry is adopted to check the effect of technique reduction on SO₂ emissions. The accumulation of labour forces and human capital may lead to technological advances and increases in per capita production efficiency, which can reduce the emissions intensity. From this perspective, technological innovation is crucial for reducing the cost of emissions mitigation. Literature on technological progress suggests that CO₂ emissions are inversely related to research intensity, technology transfer and the absorptive capacity for foreign technology. Grossman and Krueger (1995) first suggest that the technical effect is one of the two driving factors of environmental emission reduction. Other similar conclusions can be found in many works (i.e. Ang, 2009; Henderson et al., 2007). Some scholars recommend that China should increase its reliance on technological progress and development for greenhouse gas (GHG) mitigation (Minihan and Wu, 2012). Our results, however, show that the emissions reduction effect of technological progress cannot fully compensate for the deterioration impact of urbanisation or industrialisation across China’s cities. When fast economic growth and a rapid urbanisation rate overtake the improvement in efficiency, emissions abatement brought by technological progress cannot offset the increase in emissions from the industrial activity.

Regarding the environmental impact of income, we test the quadratic influence of the economic growth represented by GDP per capita. The location of a large majority of cities in 2011 is on the upward sloping part of the inverted U-shape, suggesting that rising GDP per capita has the effect of increasing emissions. The per capita GDP in the later stage is more likely to be increased by the service sectors, cleaner manufacturing and technological innovation with high value-added. Another possible reason is the rising demand for green cities when urban habitants become richer and better educated (Zheng and Kahn, 2013).

Policy implications

Our analysis offers very important policy implications through understanding the spatial distribution patterns of SO₂ emissions and their driving forces in China. A SO₂ emissions abatement policy should consider different emission patterns and the characteristics of different cities.

First, a policy is more likely to be effective in decelerating the increase in SO₂ emissions by setting emissions growth rate restriction for small and medium-sized cities in the western region as shown in Table 3. Most small and medium-sized cities experienced a very rapid growth in SO₂ emissions: the largest increase during the study period was more than 60 times. Therefore, the emissions abatement policy should be designed to control for the increase in SO₂ emissions in small and medium-sized cities. A SO₂ emissions reduction policy for large cities in emissions-concentrated regions should focus on a total emissions cap by lowering the total emissions volume and demanding the improvement of energy efficiency and production efficiency. Contrary to small and medium-sized cities, SO₂ emissions in some large cities decreased during the study period. The existence of the EKC indicates that cities with a higher average income appear to reduce their SO₂ emissions; however, the total emissions volume for some large cities remains very high. Policy makers should focus on the total emissions cap design for these cities to substantially reduce the overall emissions levels by requiring much lower energy emissions intensity and higher production efficiency. Resources can be used more efficiently in large cities with economies of scale.

Second, cities should be allowed to urbanise at different rates considering the potential environmental impact of rapid urbanisation. Increase in urbanisation level may be associated with increases in the welfare of city immigrants. Our results, however, indicate that a rapid increase in the urbanisation rate can be expected to increase SO₂ emissions, a conclusion that is similar with previous studies on emissions such as CO₂ (Sadorsky, 2014). Therefore, municipal policies should not pursue a uniform urbanisation pace on the sacrifice of emission aggravation given the severity of air pollution in Chinese cities. Urbanisation in city planning should include measures dealing with the potential negative effects such as air pollution brought by the expansion of cities and population increase. Nevertheless, it is important to remember that the increase in urbanisation level associated with benefits such as poverty reduction result from higher salary for immigrants in urban employment. It is thus necessary to adopt a variety of integrated approaches to deal with the relationship between urbanisation and the emissions.

In addition, policy on manufacturing emissions should be designed to internalise the negative externalities of industrial manufacturing emissions into its production costs. Policies such as an environmental pollution tax, financial punishment and high electricity prices for heavy pollution industries should be strictly enforced to internalise the negative externalities. At the same time, policies should provide incentives for manufacturing industries with more environmentally friendly production using renewable energy and resources.

Conclusions

Chinese cities are confronting deteriorating environmental pollution while the income of urban dwellers has dramatically increased and millions of people have been lifted out of poverty. This controversial phenomenon raises challenges for policy makers to balance development between urbanisation and environmental health by addressing the severity of environmental problems and key driving forces behind increasing pollution. This study focuses on urbanisation while a number of essential variables regarding industrialisation, energy consumption, demographics, economic development, technology progress, geographical characteristics and foreign investment are included. The result demonstrates that urbanisation is an important factor shaping SO₂ emissions while industrialisation remains a key driving force. The magnitude of urbanisation influence is relatively large and statistically significant (p < 0.05).

As for the quadratic effect of affluence on the emissions, our results demonstrate an inverted U-shaped relationship between income and SO₂ emissions. Initially, SO₂ emissions increase as income rises. The additional increase in income will lower SO₂ emissions. In addition to income, we also find that the impact of population on SO₂ emissions is non-monotone. It is likely that a larger population leads to the decrease in SO₂ emissions after the size of city population reaches a threshold or turning point.

SO₂ emissions are serious threat to the environment sustainability and human health across Chinese cities. To mitigate the emissions, it is critical to adopt a variety of integrated approaches. Our results suggest that a policy would be effective in decelerating the increase in the emissions by setting emissions growth rate restriction for small and medium-sized cities, especially cities in the western region. Policies on urbanisation promotion should be localised by considering individual cities’ characteristics and sustainable development. Cities should be allowed to urbanise at different rates considering the potential environmental impact of rapid urbanisation level. In addition to SO₂ pollution, future researches can link urbanisation to health, productivity, property value, traffic and welfare.

Footnotes

Funding

This research is financed by the National Natural Science Foundation of China (41301631), fundamental research funds for the central universities (JBK140402, JBK130148, and JBK160138), and fundamental research funds for outstanding oversea talents.

Notes

References

Ang

(2009) CO₂ emissions, research and technology transfer in China. Ecological Economics68: 2658–2665.

Antweiler

Copeland

Taylor

(2001) Is free trade good for the environment?The American Economic Review91: 877–908.

Astaraie-Imani

Kapelan

et al . (2012) Assessing the combined effects of urbanisation and climate change on the river water quality in an integrated urban wastewater system in the UK. Journal of Environmental Management112: 1–9.

Auffhammer

Carson

(2008) Forecasting the path of China’s CO₂ emissions using province-level information. Journal of Environmental Economics and Management55: 229–247.

BP (2013) BP Energy Outlook 2030. Available at: http://www.bp.com/content/dam/bp/pdf/energy-economics/energy-outlook-2015/bp-energy-outlook-booklet_2013.pdf.

Brajer

Mead

(2004) Valuing air pollution mortality in China’s cities. Urban Studies41: 1567–1585.

Chen

Ebenstein

Greenstone

et al . (2013) Evidence on the impact of sustained exposure to air pollution on life expectancy from China’s Huai river policy. Proceedings of the National Academy of Science (PNAS) 110: 12,936–12,941.

Cherniwchan

(2012) Economic growth, industrialization, and the environment. Resource and Energy Economics34: 442–467.

Cole

Neumayer

(2004) Examining the impact of demographic factors on air pollution. Population and Environment26: 5–21.

10.

Currie

Neidell

(2005) Air pollution and infant health: What can we learn from California’s recent experience?Quarterly Journal of Economics120: 1003–1030.

11.

Damania

Fredriksson

List

(2003) Trade liberalization, corruption, and environmental policy formation: Theory and evidence. Journal of Environmental Economics and Management46: 490–512.

12.

Dhakal

(2009) Urban energy use and carbon emissions from cities in China and policy implications. Energy Policy37: 4208–4219.

13.

Dietz

Rosa

(1997) Effects of population and affluence on CO₂ emissions. Proceedings of the National Academy of Sciences of the United States of America (PNAS) 94: 175–197.

14.

Ebenstein

(2012) The consequences of industrialization: Evidence from water pollution and digestive cancers in China. The Review of Economic and Statistics94: 186–201.

15.

Environmental Protection Agency (2014) US Environmental Protection Agency. Available at: https://www.epa.gov/so2-pollution.

16.

Frankel

Rose

(2005) Is trade good or bad for the environment? Sorting out the causality. The Review of Economics and Statistics87: 85–91.

17.

Gassebner

Lamla

Sturm

(2011) Determinants of pollution: What do we really know?Oxford Economic Papers63: 568–595.

18.

Grossman

Krueger

(1995) Economic growth and environment. The Quarterly Journal of Economics110: 353–377.

19.

Pan

Yan

(2012) Is economic transition harmful to China’s urban environment? Evidence from industrial air pollution in Chinese cities. Urban Studies49: 1767–1790.

20.

Wang

(2012) Economic structure, development policy and environmental quality: An empirical analysis of environmental Kuznets curves with Chinese municipal data. Ecological Economics76: 49–59.

21.

Henderson

Tochkov

Badunenko

(2007) A drive up the capital coast? Contributions to post-reform growth across Chinese provinces. Journal of Macroeconomics29: 569–594.

22.

Henderson

Wang

(2007) Urbanization and city growth: The role of institutions. Regional Science and Urban Economics37: 283–313.

23.

Lozano

Gutierrez

(2008) Non-parametric frontier approach to modelling the relationships among population, GDP, energy consumption and CO₂ emissions. Ecological Economics66: 687–699.

24.

Liu

Jiang

(2015) Is economic growth improving urbanization? A cross-regional study of China. Urban Studies52: 1883–1898.

25.

Streets

Zhang

et al . (2010) Sulfur dioxide emissions in China and sulfur trends in East Asia since 2000. Atmospheric Chemistry and Physics10: 6311–6331.

26.

Martinez

CIP

Silveira

(2013) Energy efficiency and CO₂ emissions in Swedish manufacturing industries. Energy Efficiency6: 117–133.

27.

Masiol

Agostinelli

Formenton

et al . (2014) Thirteen years of air pollution hourly monitoring in a large city: Potential sources, trends, cycles and effects of car-free days. Science of the Total Environment494–495: 84–96.

28.

Minihan

(2012) Economic structure and strategies for greenhouse gas mitigation. Energy Economics34: 350–357.

29.

Poumanyvong

Kaneko

(2010) Does urbanization lead to less energy use and lower CO₂ emissions? A cross-country analysis. Ecological Economics70: 434–444.

30.

Winchester

Karplus

et al . (2014) Will economic restructuring in China reduce trade-embodied CO₂ emissions?Energy Economics42: 204–212.

31.

Sadorsky

(2014) The effect of urbanization on CO₂ emissions in emerging economies. Energy Economics41: 147–153.

32.

Stern

(2004) The rise and fall of the Environmental Kuznets Curve. World Development32: 1419–1439.

33.

Streets

Bond

Carmichael

et al . (2003) An inventory of gaseous and primary aerosol emissions in Asia in the year 2000. Journal of Geophysical Research108(D21): 8809–8823.

34.

World Bank (2007) Cost of Pollution in China: Economic Estimates of Physical Damages. Washington, DC: World Bank.

35.

World Bank and the Development Research Center of the State Council, China (DRC) (2014) Urban China: Toward Efficient, Inclusive, and Sustainable Urbanization. Washington, DC: World Bank.

36.

World Resources Institute (2014) World Resources Institute. Available at: http://www.wri.org/our-work/topics/china.

37.

(2011) The use of a goal for SO₂ mitigation planning and management in China’s 11th Five-Year Plan. Journal of Environmental Planning and Management54: 769–783.

38.

York

Rosa

Dietz

(2003) STIRPAT, IPAT and ImPACT: Analytic tools for unpacking the driving forces of environmental impacts. Ecological Economics46: 351–365.

39.

Zhang

(2002) What explains China’s rising urbanisation in the reform era?Urban Studies39: 2301–2315.

40.

Zhang

Wang

et al . (2015) Regional differences in Chinese SO₂ emission control efficiency and policy implications. Atmospheric Chemistry and Physics15: 6521–6533.

41.

Zhang

Qiao

Zhu

(2014) Structural decomposition analysis and evaluation of the Chinese emission reduction policy: Changes in SO₂ emission from 2001 to 2010. Journal of Applied Sciences14: 573–579.

42.

Zhang

Cheng

(2009) Energy consumption, carbon emissions, and economic growth in China. Ecological Economics68: 2706–2712.

43.

Zheng

Kahn

(2013) Understanding China’s urban pollution dynamics. Journal of Economic Literature51: 731–772.

Urbanisation and the geographic concentration of industrial SO 2 emissions in China

Abstract

Keywords

Introduction

Methods

A minimalist model

Data and measurement

Results

The distribution of SO2 emissions

Statistical results

Robustness checks, discussion and policy implications

Robustness checks

Discussion

Policy implications

Conclusions

Footnotes

Funding

Notes

References

The distribution of SO₂ emissions