# Effect of Energy Development and Technological Innovation on [PM.sub.2.5] in China: A Spatial Durbin Econometric Analysis.

1. IntroductionIn the first quarter of 2013, China experienced severe hazy weather that affected 1.3 million square kilometers and 800 million people. Since then, this haze has continued to shroud most parts of China. Energy consumption, directly or indirectly, increases the degree of haze pollution. How does energy development affect haze pollution? At the same time, China is speeding up the construction of an innovative country. Studying energy development and technology innovation on haze pollution have important implications for building a beautiful China.

The main components of haze pollution are [PM.sub.2.5] and [PM.sub.10]. Particles with an aerodynamic equivalent diameter of 10[micro]m or less are called [PM.sub.10]. Particles with diameters less than 2.5[micro]m are called [PM.sub.2.5]. Energy development includes energy intensity, energy structure, and energy prices. The literature has studied the impact of energy intensity on haze pollution. For example, Cheng et al. (2017) found that China's energy intensity has a significant positive impact on [PM.sub.2.5] and should limit the rapid growth of energy intensity [1]. Liu and Jiang (2017) found that a decrease in energy intensity would reduce [PM.sub.10] concentrations in the region [2]. Regarding energy structure contamination with haze, Ma and Zhang (2014) asserted that in the long run, changing the energy consumption structure and optimizing the industrial structure are the keys to controlling haze. In the short term, reducing the use of inferior coal is an effective method [3]. Hao and Liu (2015) indicated that to reduce haze pollution in China, the energy structure in secondary industries should be improved by reducing coal consumption and increasing the consumption of clean energy, such as natural gas and renewable energy [4]. Wei and Ma (2015) pointed out that adjusting energy structure and technological progress are the fundamental means to control haze [5]. Li (2016) found that the cause of haze pollution in China was the backwardness of energy technologies, the higher proportion of coal consumption, and the greater proportion of heavy industries [6].Wang X et al. (2018) found that clean energy consumption had negative effects on [PM.sub.2.5] concentrations in China [7].

Technological innovation is the introduction of new products, services, or elements in the production process or service operations [8]. Technological innovation accelerates economic growth by increasing productivity and reduces energy consumption in the production process by increasing energy efficiency, ultimately reducing pollution emissions [9]. The literature has investigated the impact of technological innovations on atmospheric pollution. For example, Popp (2003) used patent data to study the innovation of the United States' flue gas desulfurization devices ("scrubbers") and showed that technological innovations since 1990 increased scrubber removal efficiency [10]. Shi and Lai (2013) found that the diffusion of low-carbon and green technological innovations promotes regional economic development but saves energy through low-carbon dioxide emissions or zero energy emissions [11]. Yang and Li (2017) studied the relationship between technological change and carbon emissions based on China's 1997-2010 panel data for 30 provinces and found that technological progress is the greatest contributor to CO2 emission reductions [12]. Wang et al. (2018) estimated the impact of Chinese technical knowledge on the intensity of industrial air pollutants based on the learning curve and environmental learning curve theories and found a causal relationship between technical knowledge and the intensity of industrial air pollutants [13].

The literature has also investigated the impact of technological innovations on haze. For example, Chang et al. (2017) applied regional optimization methods to analyze the impact of technological development in Shanghai on [PM.sub.2.5] pollution and found that decreased fossil energy consumption and increased use of clean energy can reduce [PM.sub.2.5] contamination [14]. Xu and Lin (2018) believe that China's energy-saving emission reduction technology continues to lag behind developed countries, which makes reducing [PM.sub.2.5] pollution in China is a difficult problem to overcome [15]. Zhang et al. (2016) suggested that technological innovation should be strengthened to prevent and control haze pollution in Beijing [16]. Liu (2018) found that technological innovation reduces [PM.sub.10] at a regional level and indirectly reduces [PM.sub.10] in adjacent regions through knowledge spillover effects [17].

The literature has also discussed the impact of energy development and technological innovation on haze pollution, but the following deficiencies have been observed. Firstly, from the perspective of energy development, the literature on the impact of energy structure on haze pollution is abundant but investigations of the impact of energy development combined with technological innovation on the haze pollution are scarce. Secondly, the literature on the impact of technological innovation on environmental pollution has been increasing, but the literature on the impact of technological innovation on haze pollution, especially on [PM.sub.2.5], is relatively scarce. Thirdly, from the perspective of research methods, most of the literature has analyzed the impact of technological innovation on haze pollution from the perspective of ordinary panels, but investigations from the perspective of spatial measurement are still lacking. Finally, there is a dearth of literature on the use of technical innovation infrastructure indicators to measure technological innovation, which is usually measured by patents or R&D input indicators.

Based on the deficiencies in the literature, this paper intends to expand on the following aspects. First, energy development and technological innovation are included in the same framework to examine the impact of energy development and technological innovation on [PM.sub.2.5]. Secondly, the spatial Durbin model with an adjacent weight matrix and distance weight matrix is used to analyze the impact of energy development and technological innovation on [PM.sub.2.5]. Finally, indicators such as technological innovation infrastructure are used to refer to technological innovation.

2. Methodology and Data

2.1. Global Spatial Autocorrelation. The global spatial autocorrelation can effectively describe the overall situation of [PM.sub.2.5] changes at the interprovincial level and is usually analyzed by the Global Moran's I index. This index was first proposed by Moran (1950) [18]. The Global Moran's I index reflects the similarities between units in each region and adjacent regions through a spatial weight matrix (Anselin, 1988) [19]. Its calculation formula is as follows:

[mathematical expression not reproducible] (1)

where [S.sup.2] = (1/n)[[summation].sup.n.sub.i=1][([x.sub.i] - [bar.x]).sup.2] is the variance of the observations. [bar.x] = (1/n)[[summation].sup.n.sub.i=1][x.sub.i] is the average of all spatial unit observations. [x.sub.i] and [x.sub.j] represent the ith and jth spatial unit observations, respectively, and n is the total number of spatial units. [w.sub.ij] is a spatial weight matrix. The range of the Global Moran's I Index is [-1, 1]. I>0 indicates a positive spatial autocorrelation of interprovincial [PM.sub.2.5] in China. The larger the value is, the stronger the agglomeration is; that is, the high value zone is adjacent to the high value zone, and the low value zone is adjacent to the low value zone. I<0 indicates a negative autocorrelation of the interprovincial [PM.sub.2.5] in China; that is, the high value zone is adjacent to the low value zone. I=0 indicates no spatial autocorrelation in the provincial [PM.sub.2.5] in China.

The significance was tested using a standardized statistic, Z, with the following formula:

Z = [I - E(I)]/[square root of VAR(I)] (2)

In (2), E(I) and VAR(I) are the expected values and variances of Moran's I index, respectively. At a significant level of 5%, if [absolute value of Z] > 1.96, there is a significant spatial autocorrelation.

Anselin (1996, 2002) [20, 21] believes that the Global Moran's I index scatter plot can show the spatial autocorrelation. The first and third quadrants illustrate the positive spatial correlation of the observations. The second and fourth quadrants describe the negative spatial correlation of observations.

2.2. Spatial Regress Models. The spatial panel model mainly includes spatial lag model SLM, spatial error model SEM, and spatial Durbin model SDM (Anselin et al., 2008) [22]. According to Elhorst (2012), the spatial lag model SLM formula is as follows [23]:

[y.sub.it] = [delta][N.summation over (j=1)][w.sub.ij][y.sub.jt] + [alpha] + [x.sub.it][beta] + [[mu].sub.i] + [[lambda].sub.t] + [[epsilon].sub.it], i = 1, ..., N, t = 1, ..., T (3)

where [y.sub.it] = [([y.sub.1t], [y.sub.2t], [y.sub.3t] ..., [y.sub.Nt]).sup.T] it is an Nx1 dimensional vector composed of explanatory variables. [[summation].sup.N.sub.j=1][w.sub.ij][y.sub.jt] is the interaction effect between the interpreted variable [y.sub.it] and the adjacent elements [y.sub.jt] and [w.sub.ij] are an NxN nonnegative spatial weight matrix. [delta] is the response parameter of the endogenous interaction effect. [alpha] is a constant item. [x.sub.it] is an N x K exogenous explanatory variable matrix. [beta] is the matching response parameter. [[epsilon].sub.it] is an independent identical distribution error term, subject to (0, [[sigma].sup.2]) distribution. [[mu].sub.i] represents the spatial effect. [[lambda].sub.t] is the time fixed effect.

This paper uses two types of spatial weight matrices. One is the traditional binary spatial adjacent matrix, hereinafter referred to as the adjacent matrix.

[mathematical expression not reproducible] (4)

The other is the distance weight. [d.sub.ij] is the distance between the geographical centers of the two regions.

[mathematical expression not reproducible] (5)

where d is the distance between the geographical center of the two regions.

The spatial error model, SEM, which describes the spatial disturbance correlation and spatial overall correlation, is given by

[y.sub.it] = [alpha] + [x.sub.it][beta] + [[mu].sub.i] + [[lambda].sub.t] + [[phi].sub.it] [[phi].sub.it] = [rho][N.summation over (i=1)][w.sub.ij][[phi].sub.it] + [[epsilon].sub.it] (6)

where [[phi].sub.it] is the spatial autocorrelation error and [rho] is the spatial error correlation coefficient which estimates the degree of influence of the error shock of adjacent units with regard to the dependent variable on the observation value of the unit. The other parameters are the same as defined above.

The spatial Durbin model, SDM, which includes the spatial lag value of the explanatory variables is given by

[y.sub.it] = [delta][N.summation over (i=1)][w.sub.ij][y.sub.jt] + [alpha] + [x.sub.it][beta] + [N.summation over (j=1)][w.sub.ij][x.sub.ijt][theta] + [[mu].sub.i] + [[lambda].sub.t] + [[epsilon].sub.it] (7)

where both [theta] and [beta] are K x 1 dimensional parameter vectors. The other parameter implication is the same as above. Due to the introduction of the spatial weight matrix, the spatial econometric may produce endogenous variables. If the ordinary least squares estimation is still used in the spatial econometric model, a bias factor will be produced; thus, the maximum likelihood method is used for estimation.

2.3. Model Specification. To analyze the impact of energy development, technological innovation, and other factors on interprovincial [PM.sub.2.5] in China, the following model was constructed:

[mathematical expression not reproducible] (8)

In formula (8), [PM.sub.2.5] is the particulate matter that can enter the lung, or fine particulate matter. Ei is the energy intensity, which is the ratio of the energy consumption of each province to the regional GDP. Cs is the energy structure, which is the proportion of coal consumption in energy consumption. Ppi is the price of energy consumption, which is represented by the producer price index of industrial producer. Npi is the sales revenue of new products of industrial enterprises above the scale. Patent is the amount of patent application authorization. Infra is technological innovation infrastructure that is characterized by internet penetration rate. Market is the turnover of the technology market.

2.4. Data. This article examines 29 provincial administrative regions in mainland China. Due to the lack of data on Tibet and Chongqing, it does not consider Tibet and Chongqing. The sources of the raw data are the 2004-2017 China Statistical Yearbook, China Energy Statistical Yearbook, China Science and Technology Statistical Yearbook, statistical yearbooks of various provinces and districts, and statistical bulletins on the national economic and social development of various provinces and districts over the years. China's monitoring of [PM.sub.2.5] started late, that is, well after the problem began. Therefore, to acquire [PM.sub.2.5] data, the annual mean value data from 2003 to 2010 was mainly based on the measurement of global aerosol optical depth from satellite-borne equipment operated by a Columbia University research team and the Battle Research Institute (Donkelaar et al. 2010) [24]. The data of the population-weighted average annual [PM.sub.2.5] for all provinces of China from 2003-2010 are also derived from these sources. [PM.sub.2.5] data from 2011-2012 were calculated based on monitoring data from the China Environmental Monitoring Center. The years 2013-2016 are from the China Statistical Yearbook (2014-2017). Referring to Ma et al. (2016), provincial capital data are used instead of provincial data [25]. The descriptive statistics of the relevant variables are shown in Table 1.

The correlation coefficient and multicollinearity test of the variables are shown in Table 2. It can be observed that, in addition to the correlation coefficient between Patent and Npi being as high as 0.9335, other correlation coefficients are lower than 0.6, and some correlation coefficients do not pass the significance test. Using the variance inflation factor VIF to test for multicollinearity, the results show that a VIF greater than 1 is less than 6.95, and the mean is 3.797. Since the VIF values are all less than 10, there is no multicollinearity.

3. Temporal and Spatial Characteristics of Interprovincial [PM.sub.2.5] in China

3.1. Temporal Variation Characteristics of the Interprovincial [PM.sub.2.5]. From the perspective of the temporal characteristics, the interprovincial [PM.sub.2.5] in China showed an inverted "N" trend, that is, a slow downward trend before 2012, an upward trend after 2012, a peak after reaching a peak in 2013, and then a downward trend. These trends are in line with China's situation. China's haze pollution showed a major increase in 2013. Since then, the country has committed to the management of haze pollution, and haze pollution has shown a downward trend.

According to the traditional eight regional division methods, the interprovincial [PM.sub.2.5] in China is divided into eight regions (Figure 1). The descriptive statistics of [PM.sub.2.5] concentrations in the eight regions are shown in Table 3. The average [PM.sub.2.5] concentrations in the northern coast and middle Yangtze River are higher than the national average. Except for 2009, the average [PM.sub.2.5] concentration in the middle Yellow River is higher than that of the whole country. The mean value of [PM.sub.2.5] in the southern coastal areas is lower than the national average. As presented in Table 3, the average annual [PM.sub.2.5] concentration in China is 37.5951[micro]g/[m.sup.3]. In the eight regions, the mean concentration of [PM.sub.2.5] in the northern coast is 54.1728, ranking first. The average concentration of [PM.sub.2.5] in the middle Yangtze River and middle Yellow River was higher than 40[micro]g/[m.sup.3], ranking second and third, respectively. The average concentration of [PM.sub.2.5] in the eastern coast ranks fourth. The mean [PM.sub.2.5] concentrations in these four regions are higher than the national average. The average concentration of [PM.sub.2.5] in the southern coastal areas is the lowest, from low to high, followed by the northwest, northeast, and southwest regions, in that order. The [PM.sub.2.5] concentrations in these four regions are lower than the national average.

3.2. Spatial Differentiation Characteristics of the Interprovincial [PM.sub.2.5]

3.2.1. Spatial Distribution Pattern of the Interprovincial [PM.sub.2.5]. Using ArcGIS10.2 software, according to the quartile method, the distribution map of [PM.sub.2.5] in China from 2003 to 2016 is plotted (Figure 2).

From low to high, the interprovincial [PM.sub.2.5] average is divided into four levels. The first-level provinces are Hainan, Fujian, Inner Mongolia, Ningxia, Heilongjiang, and Qinghai. The second-tier provinces are Guizhou, Jilin, Gansu, Yunnan, Guangdong, Zhejiang, Xinjiang, Shanghai, Liaoning, Jiangxi, and Shanxi. The third-level provinces are Guangxi, Shaanxi, Hunan, Tianjin, Beijing, Anhui, Hubei, Sichuan, and Jiangsu. The fourth level is Henan, Shandong, and Hebei. It can be observed that the mean values of [PM.sub.2.5] in the northern coast and middle Yellow River are relatively high and that of Guizhou, Yunnan, and the northeast in the south coast and southwest regions are relatively low. The interprovincial [PM.sub.2.5] in China shows a trend of high and low in the east, and there is a spatial agglomeration.

3.2.2. Global Spatial Autocorrelation Analysis of the Interprovincial [PM.sub.2.5]. By using GeoDA9.5 software and rook neighboring, Moran's I index of the interprovincial [PM.sub.2.5] average value in China from 2003 to 2016 is 0.4203. After 999 permutations, the P value is 0.0020, and the normal statistic Z value is 3.6473, which is larger than the critical value 1.96 of the normal distribution function at the 0.05 significant level and indicates that the interprovincial [PM.sub.2.5] average value spatial autocorrelation is significant.

Moran's I index scatter plot is used to further examine the state of spatial agglomeration (Figure 3). The first quadrant is the H-H agglomeration. Ten provinces and autonomous regions, namely, Hebei, Shaanxi, Shandong, Henan, Jiangsu, Beijing, Tianjin, Anhui, Hubei, and Shanxi, are in this quadrant, account for 34.48% of all the investigated provinces, and are mainly located in the north coast and middle Yellow River. The third quadrant is the L-L agglomeration. Ten provinces and autonomous regions, namely, Inner Mongolia, Fujian, Guangdong, Zhejiang, Shanghai, Gansu, Ningxia, Heilongjiang, Jilin, and Xinjiang, are in this quadrant, together with Hainan, which spans the second and third quadrants and accounts for 37.93% of all the provinces. These provinces are mainly located in the southern coast, the eastern coast, and the northeast. The second quadrant is L-H, with five provinces and regions in Jiangxi, Liaoning, Yunnan, Guizhou, and Qinghai, and Hainan. The fourth quadrant is an agglomeration of H-L, and Sichuan, Hunan, and Guangxi are located in this quadrant. In general, the proportion of provinces with the same spatial autocorrelation in the H-H and L-L quadrants is 72.41%. In the L-H quadrant and the H-L quadrant, the provinces with different spatial autocorrelations accounted for only 31.03%. Based on these results, it can be observed that the interprovincial PM2.5 regional integration in China is significant.

4. Empirical Research

Based on the spatial agglomeration of [PM.sub.2.5], the spatial econometric model is used to analyze the impact of energy development and technological innovation on [PM.sub.2.5].

4.1. Spatial Diagnostic Test. In Section 2, as aforementioned, there are generally three types of spatial panel models. To determine which model to use, a nonspatial interaction model is used for spatial diagnostic tests based on the model selection mechanism proposed by Anselin (2005) [26]. The results are shown in Table 4. For the adjacent weights, the LM-lag and LM-error tests show that both the hypothesis of no spatially lagged dependent variable and the hypothesis of no spatially auto-correlated error term must be rejected at 1% significance. The time fixed effect did not pass the robust LM-lag test. The spatial fixed effect did not pass the significance test of robust LM-error. According to Elhorst (2014) [27], the spatial Durbin panel model was chosen. The measurement method is the maximum likelihood estimation proposed by Elhorst (2003) [28].

For the distance weight, the LM-lag and LM-error tests show that both the hypothesis of no spatially lagged dependent variable and the hypothesis of no spatially auto-correlated error term must be rejected at 1% significance. The robust LM-lag test shows that the pooled OLS and time fixed effects do not pass the significance test. Similar to the robust LM-error test, only the pooled OLS passes the significance test; thus, the spatial Durbin panel model is chosen in the same way as the adjacent weights. The LR test is used to select the fixed effects. The results indicate that at the 1% level of significance the null hypothesis that the spatial fixed effects are jointly insignificant must be rejected. Similarly, the null hypothesis that the time-period fixed effects are jointly insignificant must be rejected. According to Baltagi (2005) [29], the two-way fixed effects model is selected, which is also known as the two-way fixed effects model.

4.2. Empirical Analysis. The results of the spatial Durbin model with two-way fixed effects are shown in Table 5. The Wald and LR tests of the two weights show that at the 1% level of significance, the null hypothesis of the spatial Durbin model SDM into the spatial lag model SLM is rejected, and the spatial Durbin model SDM is also rejected as the spatial error model SEM; therefore, a spatial Durbin model SDM is used. The estimated value of the Hausman test of the adjacent weight is 95.6264, the degree of freedom is 15, and the value of P is 0. Therefore, the random effect is rejected, and the two-way fixed effect model is used. Elhorst (2014) indicated that the coefficients of the spatially lagged dependent variable and independent variables are quite sensitive to bias corrections. Therefore, in the following analysis of adjacent weights, the two-way fixed effect model with bias correction is selected (the third column of data in Table 4). The estimated value of the distance weight Hausman test is 60.0659, the degree of freedom is 15, and the value of P is 0. Therefore, the model also rejects the random effect and selects the bi-directional fixed effect model. That is, distance weights are used to describe the third column of the second half of Table 5.

Comparing the estimation results in Tables 5 and 4, it can be observed that the [R.sup.2] of the noninteractive effect model is smaller whether it is the adjacent weight or the distance weight. The noninteractive effect of an adjacent weight and distance weight two-way fixed model [R.sup.2] are both 0.1661. The two-way fixed effect of the spatial Durbin model [R.sup.2] is 0.9398, and the distance weight error correction two-way fixed effect Durbin model [R.sup.2] is 0.9297. This paper illustrates that the use of a spatial Durbin model to analyze the impact of energy development and technical innovation on [PM.sub.2.5] is reasonable. The spatial auto-regression coefficients of adjacent weights and distance weights are 0.4268 and 0.5168, respectively; all pass the 1% level significance test, indicating that the [PM.sub.2.5] of adjacent regions have a positive impact on local region. A decrease of one percentage point in [PM.sub.2.5] in adjacent regions results in a decrease in [PM.sub.2.5] in this area by 0.4268, 0.5168 percentage points, respectively. [PM.sub.2.5] has a spatial spillover effect. This is consistent with the robust LM test of the spatial autocorrelation of Table 4. Additionally, the auto-regression coefficient of the distance weights is significantly larger than the adjacent weight because the spillover of [PM.sub.2.5] is related to the geographical location. Even when not geographically adjacent, a certain area affects other areas. Therefore, this paper uses two types of weight matrix, namely, the adjacent weight and distance weight, to analyze the spatial spillover effect of provincial [PM.sub.2.5] in China in order to reflect the objective reality comprehensively and accurately.

Since the significance of each variable estimation coefficient in the nonspatial model is not the same as in the spatial econometric model, the coefficient in Table 5 cannot be compared with the Table 4. LeSage and Pace (2009) believe that direct and indirect effects explain the true spatial spillover effect of each variable. Therefore, this paper decomposes the direct and indirect effects of explanatory variables. The results are shown in Table 6. Due to the feedback effect, the direct effects of the variables in Table 6 are different than in Table 5. Elhorst (2014) believes that the feedback effect is partly due to the coefficient estimate of the spatially lagged dependent variable and partly to the coefficient of the spatially lagged value of the explanatory variable itself. The decomposition result of the adjacent weight is better than the distance weight, and the coefficient signs are roughly the same. Therefore, the results of the adjacent weights are used for analysis.

The direct effect of energy intensity is positive, but it has not passed the significance test, indicating that the positive effect of energy intensity on haze pollution in the region is not obvious. The indirect effects and total effects of energy intensity are significantly positive. The increase in energy intensity will significantly increase the concentration of [PM.sub.2.5] in adjacent areas, and subsequently increase the global [PM.sub.2.5] concentration. The direct effects and total effects of energy structure are significantly positive. The increase in the proportion of coal consumption increases the [PM.sub.2.5] concentration in the region and globally. The proportion of coal consumption increase 1 percentage will increase the [PM.sub.2.5] concentration in the region by 0.1012 percentage points and increase the global [PM.sub.2.5] concentration by 0.2628 percentage points. Therefore, in order to reduce [PM.sub.2.5] concentrations, China should reduce the proportion of coal consumption.

The direct effect and total effect of energy consumption price are significantly negative. The increase in energy consumption price decreases the [PM.sub.2.5] concentration in the region and globally. The energy consumption price increase 1 percentage will reduce the concentration of [PM.sub.2.5] in the region by 1.1418 percentage points and reduce the global [PM.sub.2.5] concentration by 2.4350 percentage points. It shows that the rise of energy consumption price can effectively reduce the [PM.sub.2.5] concentration. The direct, indirect, and total effects of the sales revenue of new products of industrial enterprises are significantly negative. The increase in sales revenue of new products will reduce the concentration of [PM.sub.2.5] in the region by 0.1285 percentage points, reduce the concentration of [PM.sub.2.5] in adjacent regions by 0.3058 percentage points, and reduce the global [PM.sub.2.5] concentration by 0.4343 percentage points. It shows that the sales revenue of new products of industrial enterprises has achieved remarkable results in reducing [PM.sub.2.5] concentration. The indirect effects and total effects of patents are significantly negative. The increase in the number of patent applications granted in the region will reduce the [PM.sub.2.5] concentration of the adjacent regions and globally. The indirect effect of Infra is significantly negative, indicating that the increase in infrastructure for technological innovation in the region will reduce the concentration of [PM.sub.2.5] in adjacent areas. The direct, indirect, and total effects of the turnover of the technology market are significantly negative. The increase in the turnover of the technology market will not only reduce the concentration of [PM.sub.2.5] in the region but also the concentration of [PM.sub.2.5] in the adjacent region through the spillover effect, significantly reducing Global [PM.sub.2.5] concentration. Therefore, in order to reduce [PM.sub.2.5] concentration, it is necessary to increase the number of patent applications granted, expand technological innovation infrastructure and increase turnover of the technology market in China.

5. Conclusion and Policy Implications

This paper uses the panel data of 29 provinces and regions in China (excluding Tibet and Chongqing) to analyze the spatial and temporal characteristics of [PM.sub.2.5]. The spatial Durbin model is used to empirically examine the impact of energy development and technological innovations on [PM.sub.2.5]. The following are the main findings of this research.

From the perspective of temporal characteristics, the interprovincial [PM.sub.2.5] in China shows an inverted "N" trend. The average concentrations of [PM.sub.2.5] in the northern coast, middle Yangtze River, middle Yellow River, and eastern coast are higher than the national average. The average concentrations of [PM.sub.2.5] in the south coast, northwest, northeast, and southwest regions are lower than the national average. The interprovincial [PM.sub.2.5] in China is high in the east and low in the west, and a spatial agglomeration is observed. The [PM.sub.2.5] has a significant autocorrelation, and the regional integration trend is significant. The spatial autoregression coefficients of adjacent weights and distance weights are significantly positive, and haze pollution in adjacent regions has a positive impact on local region. There is a spatial spillover effect of interprovincial [PM.sub.2.5] in China. The indirect effects and total effects of Energy intensity are significantly positive. The increase in energy intensity will significantly increase the concentration of [PM.sub.2.5] in adjacent areas, and subsequently increase the global [PM.sub.2.5] concentration. The direct effects and total effects of Energy structure are significantly positive. The increase in the proportion of coal consumption increases the [PM.sub.2.5] concentrations in the region and globally. The direct effect and total effect of energy consumption price are significantly negative. The increase in energy consumption price reduces the [PM.sub.2.5] concentration in the region and globally. The direct, indirect, and total effects of the sales revenue of new products of industrial enterprises are significantly negative. The increase in sales revenue of new products in local region will reduce the concentration of [PM.sub.2.5] in the region by 0.1285 percentage points and that in adjacent regions by 0.3058 percentage points, which will reduce the global [PM.sub.2.5] concentration by 0.4343 percentage points. The indirect effects and total effects of patents are significantly negative. The increase in the number of patent applications granted in the region will reduce the [PM.sub.2.5] concentrations of the adjacent regions and globally. The indirect effect of Infra is significantly negative, indicating that the increase in infrastructure for technological innovation in the region will reduce the concentration of [PM.sub.2.5] in the adjacent areas. The direct, indirect, and total effects of the turnover of the technology market are significantly negative. The increase in the turnover of the technology market will not only reduce the concentration of [PM.sub.2.5] in the region but also reduce the concentration of [PM.sub.2.5] in the adjacent region through the spillover effect, significantly reducing global [PM.sub.2.5] concentration.

Based on the main findings of this study, this paper proposes the following recommendations for China. First, continue to promote the regional collaborative management of [PM.sub.2.5]. Based on the spatial spillover effect of [PM.sub.2.5], the control of [PM.sub.2.5] requires joint prevention [7]. After the outbreak of smog pollution, the Chinese government implemented strong measures to control haze pollution, including joint prevention and control of the Beijing-Tianjin-Hebei region. The inverted "N" curve of China's [PM.sub.2.5] indicates that these governance measures are effective. Additionally, continue to promote regional collaborative governance and form a joint force to control haze pollution. The concentrations of [PM.sub.2.5] in the southern coastal and southwest regions are relatively low, and these regions can work together to consolidate existing achievements and present these positive effects throughout the country. The concentrations of [PM.sub.2.5] in the northern coastal areas and the middle Yellow River are relatively high. The provinces and regions within these regions must negotiate, cogovern, work together, share resources and information, and reduce their respective concentrations of [PM.sub.2.5].

Secondly, reduce energy intensity [30]. An increase in energy intensity will increase the concentration of [PM.sub.2.5]; thus, it is necessary to increase energy efficiency and reduce energy intensity through technological innovations, that is, directly by reducing the concentration of [PM.sub.2.5] and indirectly by improving the level of technological innovation, to reduce [PM.sub.2.5] pollution. Taxation and other economic means will encourage enterprises to carry out technological transformations to improve their technological level and reduce energy intensity; research and develop clean technologies; implement clean production processes; and eradicate the generation and emission of [PM.sub.2.5] from the source.

Again, optimize energy structure and implement energy substitution. An increase in the proportion of coal consumption increases the concentration of [PM.sub.2.5] in the region and globally. Therefore, China should reduce its proportion of coal consumption and optimize its energy structure. Although the rapid reduction of China's coal consumption is a difficult task in the short term, it is possible to manage the pollution control of bulk coal by formulating coal quality standards, using gasification and purification methods to generate electricity, and employing liquid fuels to utilize coal in a cleaner manner [31]. Additionally, as soon as possible, achieve coal-fired power generation on behalf of coal while improving coal quality, reducing the use and supply of coal, optimizing its energy consumption structure, increasing the ratio of clean energy and renewable energy, implementing energy alternatives, supporting the development of new energy, and promoting the consumption of new energy.

Thirdly, increase energy prices. The consumption of energy is in line with the law of demand, that is, an increase in energy prices will reduce energy consumption and, subsequently, reduce [PM.sub.2.5] concentrations. At present, the price of energy in China is low, and this price distortion does not reflect the cost of energy. This phenomenon is one of the reasons for the high energy consumption in China. Therefore, China should completely change the phenomenon of lower energy prices and play a regulatory role in the market mechanism: the price should reflect the cost of energy use [32, 33]. Along with the overall advancement of China's resource tax reform, China must speed up the implementation of environmental protection tax legislation, establish an environmental tax system, and realize the combined effect of resource taxes, energy taxes, and environmental taxes.

Finally, improve the level of technological innovation. As mentioned, the sales revenue of new products of industrial enterprises, patent, turnover of the technology market, and other factors can directly or indirectly reduce the concentration of [PM.sub.2.5], and technological innovation is an effective method of reducing [PM.sub.2.5]. Therefore, China must seize the opportunity to build an innovative country, enhance disruptive technological innovation, and improve the level of technological innovation. China can learn from the literature on developed countries and consider effective and advanced governance technology that can be absorbed and reformed and applied to China's [PM.sub.2.5] governance [34]. Talent is the basis of technological innovation; thus, it is possible to increase technological innovation personnel through incentive measures. Efficiency improves the level of technological innovation and the capital scale of technological innovation. Technological innovation requires financial support; thus, the financing channels for technological innovation must be broadened. Additionally, expanding the scale of technological innovation capital also promotes the expansion of technological innovation and production scale.

https://doi.org/10.1155/2018/2148318

Data Availability

The data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

The author declares that they have no conflicts of interest.

Acknowledgments

This paper is sponsored by Qing Lan Project in Jiangsu Province (17EYD006), the social science fund project in Jiangsu Province, the Major Program of National Social Science Foundation of China [15ZDA053].

References

[1] Z. Cheng, L. Li, and J. Liu, "Identifying the spatial effects and driving factors of urban [PM.sub.2.5] pollution in China," Ecological Indicators, vol. 82, pp. 61-75, 2017.

[2] X. H. Liu and K. Jiang, "Analyzing effect of urbanization on haze pollution based on static and dynamic spatial panel model," Nongye Gongcheng Xuebao, vol. 33, no. 20, pp. 218-225, 2017.

[3] L. Ma and X. Zhang, "Spatial effects of regional air pollution and the impact of industrial structure," China Population Resources & Environment, vol. 7, pp. 157-164, 2014.

[4] Y. Hao and Y.-M. Liu, "The influential factors of urban PM2.5 concentrations in China: A spatial econometric analysis," Journal of Cleaner Production, vol. 112, pp. 1443-1453, 2016.

[5] W. Wei and Ma. X, "Optimal policy for energy structure adjustment and haze governance in china," China Population Resources & Environment, vol. 25, no. 7, pp. 6-14, 2015.

[6] M. Li, "Haze pollution control strategies in China from the perspective of energy conservation and emission reduction," Nature Environment and Pollution Technology, vol. 15, no. 3, pp. 887-893, 2016.

[7] X. Wang, G. Tian, D. Yang, W. Zhang, D. Lu, and Z. Liu, "Responses of PM2.5 pollution to urbanization in China," Energy Policy, vol. 123, pp. 602-610,2018.

[8] F. Damanpour and W. M. Evan, "Organizational Innovation and Performance: The Problem of "Organizational Lag"," Administrative Science Quarterly, vol. 29, no. 3, p. 392, 1984.

[9] K. Sohag, R. A. Begum, S. M. Syed Abdullah, and M. Jaafar, "Dynamics of energy use, technological innovation, economic growth and trade openness in Malaysia," Energy, vol. 90, pp. 1497-1507, 2015.

[10] D. Popp, "Pollution Control Innovations and the Clean Air Act of 1990," Journal of Policy Analysis and Management, vol. 22, no. 4, pp. 641-660, 2003.

[11] Q. Shi and X. Lai, "Identifying the underpin of green and low carbon technology innovation research: A literature review from 1994 to 2010," Technological Forecasting & Social Change, vol. 80, no. 5, pp. 839-864, 2013.

[12] L. Yang and Z. Li, "Technology advance and the carbon dioxide emission in China--Empirical research based on the rebound effect," Energy Policy, vol. 101, pp. 150-161, 2017.

[13] W. Wang, B. Yu, X. Yao, T. Niu, and C. Zhang, "Can technological learning significantly reduce industrial air pollutants intensity in China?--Based on a multi-factor environmental learning curve," Journal of Cleaner Production, vol. 185, pp. 137-147, 2018.

[14] Z. Chang, K. Pan, and H. Zhu, "Power generation system optimization with emission co-benefits analysis: A case study of Shanghai," in Advances in Energy Systems Engineering, Springer International Publishing, 2017.

[15] B. Xu and B. Lin, "What cause large regional differences in PM2.5 pollutions in China? Evidence from quantile regression model," Journal of Cleaner Production, vol. 174, pp. 447-461, 2018.

[16] H. Zhang, S. Wang, J. Hao et al., "Air pollution and control action in Beijing," Journal of Cleaner Production, vol. 112, pp. 1519-1527, 2016.

[17] X. H. Liu, "Dynamic evolution, spatial spillover effect of technological innovation and haze pollution in China," Energy & Environment, vol. 229, no. 6, pp. 968-988, 2018.

[18] P. A. Moran, "Notes on continuous stochastic phenomena," Biometrika, vol. 37, pp. 17-23,1950.

[19] D. A. Griffith and L. Anselin, "Spatial econometrics: methods and models," Economic Geography, vol. 65, no. 2, p. 160,1989.

[20] L. Anselin, "The Moran scatter plot as an ESDA tool to assess local instability in spatial association," Taylor and Francis, pp. 111-125, 1996.

[21] L. Anselin, "Under the hood issues in the specification and interpretation of spatial regression models," Agricultural Economics, vol. 27, no. 3, pp. 247-267, 2002.

[22] L. Anselin, J. Gallo, and H. Jayet, "Spatial panel econometrics," in The Econometrics of Panel Data, Springer, Berlin Heidelberg, 2008.

[23] J. P. Elhorst, "Matlab software for spatial panels," International Regional Science Review, vol. 37, no. 3, pp. 389-405, 2014.

[24] A. van Donkelaar, R. V. Martin, M. Brauer et al., "Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application," Environmental Health Perspectives, vol. 118, no. 6, pp. 847-855, 2010.

[25] L. M. Ma, S. L. Liu, and X. Zhang, "Study on haze pollution induced by energy structure and transportation: based on spatial econometric model analysis," Finance & Trade Economics, vol. 37, no. 1, pp. 147-160, 2016.

[26] L. Anselin, "Exploring spatial data with GeoDa: a workbook," http://geodacenter.asu.edu/system/files/geodaworkbook.pdf, 2005.

[27] J. P. Elhorst, Spatial econometrics, Springer Briefs in Regional Science, Springer, Heidelberg, 2014.

[28] J. P. Elhorst, "Specification and estimation of spatial panel data models," International Regional Science Review, vol. 26, no. 3, pp. 244-268, 2003.

[29] Baltagi. B. H, Econometric Analysis of Panel Data, vol. 7, Wiley, New York, Chichester, 3rd edition, 2015.

[30] K. Luo, G. Li, C. Fang, and S. Sun, "PM2.5 mitigation in China: Socioeconomic determinants of concentrations and differential control policies," Journal of Environmental Management, vol. 213, pp. 47-55, 2018.

[31] G. Du, S. Liu, N. Lei, and Y. Huang, "A test of environmental Kuznets curve for haze pollution in China: Evidence from the penal data of 27 capital cities," Journal of Cleaner Production, vol. 205, pp. 821-827, 2018.

[32] D.-X. Niu, Z.-Y. Song, and X.-L. Xiao, "Electric power substitution for coal in China: Status quo and SWOT analysis," Renewable & Sustainable Energy Reviews, vol. 70, pp. 610-622, 2017.

[33] K. Cheng, "Spatial overflow effect of haze pollution in China and its influencing factors," Nature Environment and Pollution Technology, vol. 15, no. 4, pp. 1409-1416, 2016.

[34] J. Chen, C. Zhou, S. Wang, and J. Hu, "Identifyingthe socioeconomic determinants of population exposure to particulate matter ([PM.sub.2.5]) in China using geographically weighted regression modeling," Environmental Pollution, vol. 241, pp. 494-503,2018.

Xiaohong Liu (iD)

Doctor of Management, Environmental Economics, Business College of Nanjing Xiaozhuang University, No. 3601, Hongjing Road, Jiangning District, Nanjing, Jiangsu, 211171, China

Correspondence should be addressed to Xiaohong Liu; amylxhong@163.com

Received 18 September 2018; Revised 17 November 2018; Accepted 29 November 2018; Published 16 December 2018

Academic Editor: Alicia Cordero

Caption: Figure 1: Time change trend of [PM.sub.2.5] in China.

Caption: Figure 2: Spatial distribution of [PM.sub.2.5] average value in China.

Caption: Figure 3: Moran's I index scatter plot of [PM.sub.2.5] average in China.

Table 1: Descriptions of all variables in econometric model. Variable Unit Mean Maximum [PM.sub.2.5] [micro]g/[m.sup.3] 37.5951 154.00 Ei Tce/[10.sup.4] Yuan 1.194 4.520 Cs % 72.776 4518.000 Ppi No unit 101.991 125.250 Npi [10.sup.4] Yuan 27244930 2.87E+08 Patent piece 24258.41 269944.0 Infra % 30.420 77.830 Market [10.sup.4] Yuan 1518325 39409752 Variable Minimum Std. Dev. Observation [PM.sub.2.5] 2.170 22.5786 406 Ei 0.270 0.705 406 Cs 9.810 221.860 406 Ppi 82.410 6.185 406 Npi 38551.00 44506476 406 Patent 70.000 45273.93 406 Infra 2.150 19.767 406 Market 1885.000 4029395 406 Table 2: Correlations analysis and VIF tests. VIF [PM.sub.2.5] Ei Cs [PM.sub.2.5] -- 1.000 Ei 3.7015 -0.3579 *** 1.000 Cs 1.0148 -0.0440 0.1386 *** 1.000 Ppi 1.4093 -0.4122 *** 0.3714 *** 0.0706 Npi 6.4635 0.3280 *** -0.4463 *** -0.0366 Patent 6.9443 0.2866 *** -0.4200 *** -0.0364 Infra 5.6643 0.4299 *** -0.5900 *** -0.0857 * Market 1.3815 0.2702 *** -0.3164 *** -0.0488 Ppi Npi Patent Infra [PM.sub.2.5] Ei Cs Ppi 1.000 Npi -0.2206 *** 1.000 Patent -0.2164 *** 0.9335 *** 1.000 Infra -0.5087 *** 0.5298 *** 0.5305 *** 1.000 Market -0.1736 *** 0.2619 *** 0.3321 *** 0.4799 *** Market [PM.sub.2.5] Ei Cs Ppi Npi Patent Infra Market 1.000 Note. *, *** indicates significance at 10% and 1% levels, respectively. Table 3: Descriptions of [PM.sub.2.5]. Region Mean Maximum Minimum Std. Dev. Northeast 29.9595 81.000 7.530 24.8257 North Coast 54.1728 154.000 27.700 27.6504 East Coast 39.3197 78.000 21.510 16.2013 Middle Yellow River 40.6401 108.000 11.430 24.7028 Middle Yangtze River 44.1114 94.000 26.870 17.2208 South Coast 20.6838 53.000 2.170 12.9610 Northwest 29.0419 88.000 12.530 20.8145 Southwest 37.1357 96.000 19.360 14.0458 Region Observations Northeast 42 North Coast 56 East Coast 42 Middle Yellow River 56 Middle Yangtze River 56 South Coast 42 Northwest 56 Southwest 56 Table 4: Estimation results of nonspatial panel model. Variables Pooled OLS Adjacent weight Intercept 11.5064 *** (0.0000) ln Ei 0.3473 *** (0.0001) ln Cs 0.2146 *** (0.0037) ln Ppi -2.5530 *** (0.0000) ln Npi -0.0179 (0.6843) ln Patent 0.2509 *** (0.0000) ln Infra -0.0352(0.4382) ln Market 0.0776 *** (0.0017) [R.sup.2] 0.4091 LM-lag 201.7888 *** (0.000) Robust LM-lag 4.6368** (0.031) LM-error 219.8807 *** (0.000) Robust LM-error 22.7288 *** (0.000) Distance weight Intercept 11.5064 *** (0.0000) ln Ei 0.3473 *** (0.0001) ln Cs 0.2146 *** (0.0037) ln Ppi -2.5530 *** (0.0000) ln Npi -0.0179 (0.6843) ln Patent 0.2509 *** (0.0000) ln Infra -0.0352 (0.4382) ln Market 0.0776 *** (0.0017) [R.sup.2] 0.4091 LM-lag 102.2239 *** (0.000) Robust LM-lag 0.6919 (0.406) LM-error 136.6806 *** (0.0000) Robust LM-error 35.1486 *** (0.0000) Fixed effects LR-test Spatial fixed effects Time-period fixed effects Variables Spatial fixed effects Adjacent weight Intercept -- ln Ei -1.1951 *** (0.0000) ln Cs 0.1257 * (0.0722) ln Ppi -2.3159 *** (0.0000) ln Npi -0.1685 *** (0.0009) ln Patent 0.2093 *** (0.0015) ln Infra -0.2274 *** (0.0007) ln Market -0.0598 (0.1106) [R.sup.2] 0.4705 LM-lag 340.0601 *** (0.000) Robust LM-lag 36.9528 *** (0.000) LM-error 303.1079 *** (0.000) Robust LM-error 0.0006 (0.980) Distance weight Intercept -- ln Ei -1.1951 *** (0.0000) ln Cs 0.1257 * (0.0722) ln Ppi -2.3159 *** (0.0000) ln Npi -0.1685 *** (0.0009) ln Patent 0.2093 *** (0.0015) ln Infra -0.2274 *** (0.0007) ln Market -0.0598 (0.1106) [R.sup.2] 0.4705 LM-lag 323.1789 *** (0.000) Robust LM-lag 46.2787 *** (0.0000) LM-error 276.9055 *** (0.0000) Robust LM-error 0.0053 (0.942) Statistics LR-test 538.5783 441.7073 Variables Time-period fixed effects Adjacent weight Intercept -- ln Ei 0.5160 *** (0.0000) ln Cs 0.0781 (0.1749) ln Ppi -1.1728 * (0.0781) ln Npi 0.1261 *** (0.0006) ln Patent 0.1263 *** (0.0024) ln Infra -0.3931 *** (0.0000) ln Market 0.1053 *** (0.0000) [R.sup.2] 0.4592 LM-lag 70.4194 *** (0.000) Robust LM-lag 1.2878 (0.256) LM-error 82.2136 *** (0.000) Robust LM-error 13.0820 *** (0.000) Distance weight Intercept -- ln Ei 0.5160 *** (0.0000) ln Cs 0.0781 *** (0.1749) ln Ppi -1.1728 *** (0.0000) ln Npi 0.1261 *** (0.0006) ln Patent 0.1263 *** (0.0024) ln Infra -0.3931 *** (0.0000) ln Market 0.1053 *** (0.0000) [R.sup.2] 0.4592 LM-lag 13.6513 *** (0.000) Robust LM-lag 1.1820 (0.277) LM-error 12.9372 *** (0.0000) Robust LM-error 0.4678 (0.0494) DOF LR-test 29 14 Variables two-way fixed effects Adjacent weight Intercept -- ln Ei 0.3958 *** (0.0078) ln Cs 0.0851 ** (0.0395) ln Ppi -1.2583 *** (0.0009) ln Npi -0.1664 *** (0.0000) ln Patent -0.0204 (0.6597) ln Infra 0.0364 (0.5603) ln Market -0.0565 ** (0.0112) [R.sup.2] 0.1661 LM-lag 66.6052 *** (0.0000) Robust LM-lag 60.7071 *** (0.0000) LM-error 39.3528 *** (0.0000) Robust LM-error 33.4547 *** (0.0000) Distance weight Intercept -- ln Ei 0.3958 *** (0.0078) ln Cs 0.0851 ** (0.0395) ln Ppi -1.2583 *** (0.0009) ln Npi -0.1664 *** (0.0000) ln Patent -0.0204 (0.6597) ln Infra 0.0364 (0.5603) ln Market -0.0565 ** (0.0112) [R.sup.2] 0.1661 LM-lag 35.8038 *** (0.000) Robust LM-lag 10.4315 *** (0.001) LM-error 27.6293 *** (0.000) Robust LM-error 2.2570 (0.133) P value LR-test 0.0000 0.0000 Note. *, **, *** indicates significance at 10%, 5%, and 1% levels, respectively. Table 5: Estimation results of spatial Durbin model. Variables Two-way fixed effects Adjacent weight [delta] 0.3679 *** (0.0000) ln Ei 0.1058 (0.4343) ln Cs 0.0899 *** (0.0089) ln Ppi -1.0526 *** (0.0018) ln Npi -0.1079 *** (0.0000) ln Patent 0.0626 (0.1363) ln Infra 0.0410 (0.4668) ln Market -0.0452 ** (0.0170) W * ln Ei 1.0147 *** (0.0004) W * ln Cs 0.0696 (0.2970) W * ln Ppi -0.4512 (0.4771) W * ln Npi -0.1558 *** (0.0076) W * ln Patent -0.3095 *** (0.0000) W * ln Infra -0.2437 ** (0.0474) W * ln Market -0.0869 ** (0.0252) Teta -- [R.sup.2] 0.9391 log-likelihood 141.4070 Wald spatial lag 71.8978 *** (0.000) LR spatial lag 67.7774 *** (0.000) Wald spatial error 87.6175 *** (0.000) LR spatial error 81.6844 *** (0.000) Hausman test Statistics 95.6264 Distance weight [DELTA] 0.4449 *** (0.0000) ln Ei 0.4862 *** (0.0000) ln Cs 0.0752 ** (0.0421) ln Ppi -0.8146 ** (0.0282) ln Npi -0.1472 *** (0.0000) ln Patent 0.0152 (0.7189) ln Infra 0.0163 (0.7785) ln Market -0.0591 *** (0.0006) W * ln Ei -0.4203 (0.2145) W * ln Cs 0.3411 * (0.0764) W * ln Ppi -1.0727 (0.1067) W * ln Npi -0.2430 *** (0.0013) W * ln Patent -0.2628 ** (0.0210) W * ln Infra -0.2226 (0.1075) W * ln Market -0.0024 (0.9589) teta -- [R.sup.2] 0.9287 log-likelihood 105.4847 Wald spatial lag 24.3571 *** (0.0000) LR spatial lag 23.2803 *** (0.0015) Wald spatial error 33.2143 *** (0.0000) LR spatial error 30.2017 *** (0.0000) Hausman test Statistics 60.0659 Variables Two-way fixed effects (with bias correction) Adjacent weight [delta] 0.4268 *** (0.0000) ln Ei 0.0980 (0.4896) ln Cs 0.0880 ** (0.0147) ln Ppi -1.0389 *** (0.0003) ln Npi -0.1043 *** (0.0002) ln Patent 0.0657 (0.1356) ln Infra 0.0454 (0.4419) ln Market -0.0439 ** (0.0000) W * ln Ei 0.9882 *** (0.0010) W * ln Cs 0.0608 (0.3844) W * ln Ppi -0.3386 (0.6104) W * ln Npi -0.1414 ** (0.0206) W * ln Patent -0.3052 * (0.0000) W * ln Infra -0.2467 * (0.0557) W * ln Market -0.0802 ** (0.0486) Teta -- [R.sup.2] 0.9398 log-likelihood 141.4070 Wald spatial lag 61.5451 *** (0.000) LR spatial lag 67.7774 *** (0.000) Wald spatial error 75.3315 *** (0.000) LR spatial error 81.6844 *** (0.000) Hausman test DOF 15 Distance weight [DELTA] 0.5168 *** (0.0000) ln Ei 0.4867 *** (0.0011) ln Cs 0.0737 * (0.0570) ln Ppi -0.7859 ** (0.0431) ln Npi -0.1441 *** (0.0000) ln Patent 0.0160 (0.7184) ln Infra 0.0203 (0.7379) ln Market -0.0579 *** (0.0057) W * ln Ei -0.4437 (0.2106) W * ln Cs 0.3277 (0.1039) W * ln Ppi -0.9562 (0.1689) W * ln Npi -0.2238 *** (0.0046) W * ln Patent -0.2619 ** (0.0280) W * ln Infra -0.2215 (0.1261) W * ln Market 0.0021 (0.9653) teta -- [R.sup.2] 0.9297 log-likelihood 105.4847 Wald spatial lag 20.5623 *** (0.0045) LR spatial lag 23.2803 *** (0.0015) Wald spatial error 28.4091 *** (0.0000) LR spatial error 30.2017 *** (0.0000) Hausman test DOF 15 Variables Random spatial effects and time-period fixed effects Adjacent weight [delta] 0.4329 *** (0.0000) ln Ei 0.0411 (0.7629) ln Cs 0.0882 ** (0.0143) ln Ppi -1.0866 *** (0.0021) ln Npi -0.0893 *** (0.0015) ln Patent 0.0905 ** (0.0310) ln Infra 0.0467 (0.4255) ln Market -00471 ** (0.0161) W * ln Ei 0.4913 * (0.0676) W * ln Cs 0.0434 (0.5334) W * ln Ppi -0.5460 (0.4074) W * ln Npi -0.0786 (0.1850) W * ln Patent -0.2093 *** (0.0016) W * ln Infra -0.2543 ** (0.0463) W * ln Market -0.0864 ** (0.0312) Teta 0.0701 *** (0.0000) [R.sup.2] 0.8902 log-likelihood -840.5862 Wald spatial lag 46.1114 *** (0.000) LR spatial lag -- Wald spatial error 54.5307 *** (0.000) LR spatial error -- Hausman test P value 0.0000 Distance weight [DELTA] 0.4399 *** (0.0000) ln Ei 0.3066 ** (0.0245) ln Cs 0.0795 ** (0.0426) ln Ppi -1.1188 *** (0.0030) ln Npi -0.1198 *** (0.0000) ln Patent 0.0646 (0.1287) ln Infra 0.0219 (0.7171) ln Market -0.0492 ** (0.0187) W * ln Ei -0.5212 * (0.0864) W * ln Cs 0.3314 (0.1030) W * ln Ppi -0.0258 (0.9525) W * ln Npi -0.1340 * (0.0685) W * ln Patent -0.0715 (0.5086) W * ln Infra -0.0901 (0.5263) W * ln Market -0.0085 (0.8572) teta 0.0782 *** (0.0000) [R.sup.2] 0.8679 log-likelihood -649.7897 Wald spatial lag 9.7989 (0.2003) LR spatial lag -- Wald spatial error 14.7569 ** (0.0392) LR spatial error -- Hausman test P value 0.0000 Note. *, **, *** indicates significance at 10%, 5%, and 1% levels, respectively. Table 6: Decomposition estimates of the direct, indirect, and total effects. Variables Direct effect Indirect effect Distance weight ln Ei 0.2323 (0.1155) 1.6573 *** (0.0010) ln Cs 0.1012 ** (0.0158) 0.1616 (0.1815) ln Ppi -1.1418 *** (0.0062) -1.2932 (0.2219) ln Npi -0.1285 *** (0.0003) -0.3058 *** (0.0055) ln Patent 0.0270 (0.5692) -0.4464 *** (0.0006) ln Infra 0.0153 (0.7962) -0.3766 * (0.0806) ln Market -0.0560 ** (0.0183) -0.1605 ** (0.0250) Adjacent weight ln Ei 0.4738 *** (0.0059) -0.3154 (0.6591) ln Cs 0.1243 ** (0.0180) 0.7165 * (0.0894) ln Ppi -0.9929 ** (0.0159) -2.6540 ** (0.0396) ln Npi -0.1819 *** (0.0000) -0.5782 *** (0.0011) ln Patent -0.0189 (0.6949) -0.5025 ** (0.0443) ln Infra -0.0126 (0.8562) -0.4191 (0.1590) ln Market -0.0595 ** (0.0164) -0.0442 (0.6603) Variables Total effect Distance weight ln Ei 1.8896 *** (0.0006) ln Cs 0.2628 * (0.0718) ln Ppi -2.4350 * (0.0507) ln Npi -0.4343 *** (0.0012) ln Patent -0.4194 *** (0.0040) ln Infra -0.3613 (0.1181) ln Market -0.2165 ** (0.0136) Adjacent weight ln Ei 0.1584 (0.8446) ln Cs 0.8408 * (0.0663) ln Ppi -3.6469 ** (0.0114) ln Npi -0.7601 *** (0.0001) ln Patent -0.5214 * (0.0559) ln Infra -0.4317 (0.2006) ln Market -0.1037 (0.3700) Note. *, **, *** indicates significance at 10%, 5%, and 1% levels, respectively.

Printer friendly Cite/link Email Feedback | |

Title Annotation: | Research Article |
---|---|

Author: | Liu, Xiaohong |

Publication: | Discrete Dynamics in Nature and Society |

Geographic Code: | 9CHIN |

Date: | Jan 1, 2018 |

Words: | 9255 |

Previous Article: | Pullback-Forward Dynamics for Damped Schrodinger Equations with Time-Dependent Forcing. |

Next Article: | Recent Trends in Dynamical Systems on Time Scales. |

Topics: |