Does economic growth and energy consumption drive environmental degradation in China’s 31 provinces? New evidence from a spatial econometric perspective

ABSTRACT

The panel data analysis points to economic and social factors contributing to NO_x, PM_2.5, PM₁₀, SO₂, and VOCs in China’s 31 provinces. The spatial correlation analysis using Global and Local Moran’s I values indicates the existence of a significant and positive spatial autocorrelation with respect to environment, economy and energy, and the high spatial correlation is evident in the eastern region, covering the northern part of Yangtze River Delta, Huaihai Economic Zone, and the lower reaches of the Yellow River Economic Belt. The empirical estimation is performed through spatial lag and spatial Durbin models. All emitted air pollutants in 31 provinces have significant spatial dependence and strong spillover effects. There is an inverted U-shaped relationship between emitted air pollutants (NO_x, PM₁₀, VOCs, and PM_2.5) and per capita GDP, which follows the EKC hypothesis. The relationship between SO₂ and per capita GDP does not follow the EKC hypothesis. There is a positive relationship between pollutant emissions and coal consumption, which is consistent with current studies for various countries like Canada, Denmark, UK and US and regions like New York State. However, the effects of science and technology investment on air pollutants are mostly positive, which is not as policy expected.

KEYWORDS:

• Emitted air pollutants
• spatial econometrics
• energy consumption
• spill over effect

JEL CLASSIFICATION:

• Q53
• Q51

Disclosure statement

No potential conflict of interest was reported by the authors.

1. Definition for Global and local spatial autocorrelation

(1) $\mathrm{M}\mathrm{o}\mathrm{r}\mathrm{a}{\mathrm{n}}^{\prime }\mathrm{s}I=\frac{n}{\sum _{i=1}^n{\left(x_i-\bar{x}\right)}^2}\frac{\sum _{i=1}^n\sum _{j=1}^n{w}_{ij}\left(x_i-\bar{x}\right)\left(x_j-\bar{x}\right)}{\sum _{i=1}^n\sum _{j=1}^n{w}_{ij}}$(1)

(2) ${\mathrm{I}}_i=\frac{n^2}{\sum {}_{i=1}^n{\left(x_i-\bar{x}\right)}^2}\frac{\left(x_i-\bar{x}\right)\sum _{i=1}^n\sum _{j=1}^n{w}_{ij}\left(x_j-\bar{x}\right)}{\sum _{i=1}^n\sum _{j=1}^n{w}_{ij}}$(2)

Global Moran’s I in Equation.(1) is used to test for overall global spatial autocorrelation, and the local Moran’s ${\mathrm{I}}_i$ in Equation.(2) describes the spatial agglomeration of emitted air pollutants between each geographic unit and its surrounding units. Where $w_{ij}$ denotes the element of the spatial weight matrix W based on a spatial adjacency relationship; if region i and j are adjacent, $w_{ij}$ is 1, otherwise $w_{ij}$ is 0; $x_i$ and $x_j$ respectively represent the emitted air pollutants for region i and j; n is the number of units; Moran’s I index takes values on the interval in [−1, 1]. This indicates that units tend toward a positive spatial autocorrelation if this index is over 0; otherwise, the negative spatial autocorrelation exists; there is no spatial correlation if the index is 0.

2. Introduction to spatial panel regression models

The econometric model in this study is expressed as follows:

(3) $lnA{P}_{it}={\alpha }_0+{\alpha }_{1}lnGD{P}_{it}+{\alpha }_2{\left(lnGD{P}_{it}\right)}^2+{\alpha }_{3}lnE{M}_{it}+{\alpha }_4{X}_{it}+{\epsilon }_{it}$(3)

where subscripts i and t respectively denote region and year; $A{P}_{it}$ as dependent variable includes NO_x, PM_2.5, PM₁₀, SO₂, and VOCs; $GD{P}_{it}$ stands for economic development; $E{M}_{it}$ represents energy mix; and $X_{it}$ represents a vector of control variables including industrial structure, foreign direct investment, traffic intensity, forest cover, and science and technology investment.

Spatial econometric models are extended from a linear regression model to reflect spatial interaction effects for each location and its neighbouring observations. In this model, the dependent variable from neighbouring regions may interact with each other; the independent variables of neighbouring regions may interact with each other. Generally, SLM, spatial error model (SEM), and SDM are frequently used to examine and measure possible interaction effects (Elhorst 2010). The SLM and SEM incorporate endogenous interaction effects among the dependent variable and interaction effects among the error terms, respectively; whereas the SDM contains a spatial lag of dependent and explanatory variables (Minguez, Montero, and Fernandez-Aviles 2013). The equations of the SLM, SEM and SDM are used in this study as shown in Equations (4)-(6), respectively.

(4) $\begin{array}{rl}& lnA{P}_{it}=\rho WlnA{P}_{it}+{\alpha }_0+{\alpha }_{1}lnGD{P}_{it}\\ & +{\alpha }_2{\left(lnGD{P}_{it}\right)}^2+{\alpha }_{3}lnE{M}_{it}+{\alpha }_4{X}_{it}+{\epsilon }_{it}{\epsilon }_{it}\sim N\left(0,{\sigma }_{it}^2\right)\end{array}$(4)

(5) $\begin{array}{rl}& lnA{P}_{it}={\alpha }_0+{\alpha }_{1}lnGD{P}_{it}+{\alpha }_2{\left(lnGD{P}_{it}\right)}^2\\ & +{\alpha }_{3}lnE{M}_{it}+{\alpha }_4{X}_{it}+{\epsilon }_{it}\\ & {\epsilon }_{it}=\lambda W{\epsilon }_{it}+u_{it},u_{it}\sim N\left(0,{\sigma }_{it}^2\right)\end{array}$(5)

(6) $\begin{array}{c}\mathrm{l}\mathrm{n}\end{array}A{P_i}_t=\rho W\mathrm{l}\mathrm{n}A{P_i}_t+{\alpha }_0+{\alpha }_1\mathrm{l}\mathrm{n}GDP{{ }_i}_t+{\alpha }_2(\mathrm{l}\mathrm{n}GDP{{}_i}_t{)}^2+{\alpha }_3\mathrm{l}\mathrm{n}E{M_i}_t+{\alpha }_4{X_i}_t+{\theta }_{1}W\mathrm{l}\mathrm{n}GD{P_i}_t+{\theta }_{2}W(lnGD{P_i}_t{)}^2+{\theta }_{3}WlnE{M_i}_t+{\theta }_{4}W{X_i}_t+{{\epsilon }_i}_t{{\epsilon }_i}_t\sim N(0,{{\sigma }_i}_t{ }^2)$(6)

where $\rho$ is the spatial autocorrelation coefficient; $WlnA{P}_{it}$ denotes the spatial lag variable; $W$ denotes the $N\times N$ spatial weight matrix; $\lambda$ denotes the spatial error autoregressive coefficient; $u_{it}$ is a vector of random errors representing normal distribution; and $\theta$ represents the spatial autocorrelation coefficient of explanation variables.

3. Selection of spatial econometrics models

The Hausman test shows that fixed effects are superior to random effects for all emitted air pollutants. Likelihood Ratio (LR) results for all emitted air pollutants indicate that the models should be extended with spatial and time-period fixed effects. Lagrange multiplier (LM) tests and robust LM tests found that with spatial and time-period fixed effects, the SLM model is suitable for NO_x, PM₁₀, and VOCs (p<0.01, as shown in Appendix A1). The results for PM_2.5 and SO₂ cannot reject the hypotheses that no spatially lagged dependent variable and no spatial autocorrelation error tem at 1% significance. Thus, we adopt the SDM and conduct Likelihood Ratio (LR) and Walds test to determine whether this model is suitable. In Table 2, probabilities for PM_2.5 and SO₂ in Wald and LR tests are all less than 0.01, indicating that the SDM is a suitable model.

Appendix A1. Tests for choosing suitable models.

In general, the SLM with spatial and time-period fixed effects is used for NO_x, PM₁₀, and VOCs, whereas the SDM with spatial and time-period fixed effects is used for PM_2.5 and SO₂.

Notes

¹ More details: http://www.mee.gov.cn/xxgk/hjyw/201301/t20130125_245527.shtml.

² The definition for global spatial autocorrelation can be found in Appendix.

³ The definition for local spatial autocorrelation can be found in Appendix.

⁴ The definition and selection for econometric models in this study are introduced in Appendix.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China grant no. [71573253]; Key Projects of Philosophy and Social Sciences for Universities by Jiangsu Provincial Department of Education grant no. [2018SJZDI109]; China Scholarship Council grant no. [201706420071].