The Energy Conservation and Emission Reduction Effects of Economic Agglomeration: A Spatial Perspective Based on China's Province-level Data

5 Based on the data of 30 provinces in China from 1995 to 2017, this paper combines exploratory spatial 6 data analysis method, dynamic spatial Durbin model, and intermediary effect model to explore the 7 spatial influence mechanism between economic agglomeration, energy intensity, and carbon emission 8 intensity. The research results provide a basis for China's early realization of energy conservation and 9 emission reduction goals, economical green development, and regional development strategy selection. 10 Firstly, the results show that China's carbon emission intensity has apparent spatial agglomeration and 11 path dependence characteristics. Secondly, the economic agglomeration has the dual effect of energy 12 saving and emission reduction. Furthermore, there is a significant inverted N - curve relationship 13 between economic agglomeration and carbon emission intensity and carbon emissions, and a 14 significant U - shaped curve relationship exists between economic agglomeration and energy intensity. 15 Finally, economic agglomeration can indirectly affect carbon emission intensity through the mediating 16 effect of energy intensity, and there is a significant inverted U - shaped curve relationship between 17 energy intensity and carbon emission intensity. Therefore, promoting mutual coordination of 18 environmental policies and building a regional collaborative governance mechanism is an effective way 19 to achieve a win - win situation for the environment and economy of Beautiful China. 20


25
Against the background of global warming, countries worldwide reduce greenhouse gas emissions 26 through global agreements, and China is facing tremendous pressure to reduce carbon dioxide 27 emissions. In 2020, China clearly stated at the United Nations General Assembly that carbon dioxide 28 emissions should peak before 2030 and strive to achieve carbon neutrality by 2060. The carbon 29 emission intensity index reflects the carbon emission efficiency in economic development, that is, the 30 carbon dioxide emission caused by unit GDP growth. At present, the research on carbon emissions has 31 been relatively sufficient, and the intensity of carbon emissions can better reflect the cost of carbon 32 emissions (Shao et al., 2018;Zhou et al., 2018). Therefore, it is essential to study China's carbon 33 emissions by paying attention to China's carbon emissions intensity. According to Figure 1, since 1995, 34 although China's carbon emission intensity has shown a downward trend as a whole, the total emission 35 intensity is still relatively large. Therefore, to achieve the goal of carbon neutrality, it is particularly 36 critical to pay attention to the influencing factors of carbon emission intensity. 37 have been implemented. Economic agglomeration improves resource utilization efficiency through 45 technology spillover effects, and at the same time, will lead to an increase in energy consumption and 46 carbon emissions, which will have a dual impact on energy conservation and emission reduction effects. 47 According to the theory of external economics (Marshall, 1920) and new economic geography (Fujita 48 et al., 1999), economic agglomeration brings positive externalities to the environment through 49 technology spillovers and economies of scale. Enterprises can share energy-saving equipment and 50 pollution control services (Xie and Yuan, 2016) to promote resource reuse (Ehrenfeld, 2010; Zhang and 51 Dou, 2013; Li and Zhang, 2013). On the other hand, economic agglomeration will lead to excessive 52 concentration of production factors, cause crowding effects, and accelerate excessive consumption of 53 resources (Frank et al., 2001; Verhoef and Nijkamp, 2002;Zeng, 2008). Furthermore, there is an 54 uncertain or nonlinear relationship between economic agglomeration and environmental pollution (Yan 55 et al., 2011;Liu et al., 2018;Zhang, 2018). Therefore, to promote the coordinated development of 56 economic agglomeration and carbon emission reduction policies and achieve a win-win effect of 57 energy conservation and emission reduction, it is particularly critical to analyze the internal impact 58 mechanism of economic agglomeration on carbon emissions. 59 The burning of fossil fuels represented by coal will directly produce pollutants such as carbon dioxide 60 and sulfur dioxide in the production and utilization process. Therefore, energy consumption is a 61 fundamental cause of environmental pollution, such as increased carbon emissions ( At present, research on carbon emissions mainly focuses on the influencing factors and control 72 methods of total carbon emissions, and the internal relationship between carbon emissions intensity and 73 economic agglomeration is rarely measured by including energy consumption intensity. In addition, the 74 main research methods are decomposition analysis and standard econometric models, which can only 75 obtain the contribution degree of influencing factors to carbon emissions but cannot explore the spatial 76 correlation of carbon emissions and the spillover effects of influencing factors on carbon emissions. 77 Since economic phenomena show temporal correlation and show spatial correlation to a certain extent, 78 it is necessary and feasible to use spatial measurement methods to explore the relationship between 79 carbon emission intensity and regional development. Therefore, focusing on China's 30 provincial-level 80 data from 1995 to 2017, the dynamic spatial Durbin model and the intermediary effect model are used 81 to analyze the relationship between economic agglomeration, energy intensity, and carbon emission 82 intensity. From the perspective of spatial economic agglomeration, this article provides a necessary 83 decision-making basis for China to effectively achieve energy conservation and emission reduction 84 goals, economical green transformation and development, and regional development strategies. 85 (2)

Materials and methods
Where 2 indicates carbon emission, indicates energy fuel, indicates energy consumption, 95 indicates an average low calorific value of energy, indicates carbon emission factor of 96 energy. The formula is as follows: 97 Where represents carbon content in energy, represents carbon oxidation factor of energy. 98 The data comes from the "China Energy Statistical Yearbook" in 1996-2018. 99

Estimation of economic agglomeration 100
Economic agglomeration mainly refers to the density of economic activities in the unit space. The 101 output density reflects the spatial distribution of economic activities and the carrying capacity of 102 economic activities per unit area, which conforms to the density characteristics of economic 103 agglomeration (Ciccone and Hall, 1993). Therefore, this paper chooses to use non-agricultural products 104 per unit area to measure the degree of economic agglomeration. 105

Explanatory variables 106
According to the STIRPAT model and the environmental Kuznets curve hypothesis, this paper sets the 107 control variables, which are population (POP), per capita income (LY), energy consumption structure 108 (ES), industrial structure (IS), technological progress (RD), and economic opening rate (FDI). The 109 specific description of each variable is shown in Table 1. 110

Exploratory spatial data analysis 119
The Moran's I index can explain the spatial correlation of the carbon emission intensity of 30 provinces 120 across the country very well, and the formula is as follows. 121 indicates sample variance, X i and X j respectively represent carbon intensity 122 of regions i and j, n is the total number of regions, W ij is the spatial weight matrix. 123 The spatial correlation of data is a prerequisite for building a spatial model, so the spatial correlation of 124 core variables needs to be tested first. According to Table 2, the results show that the Moran's I index 125 of China's carbon emission intensity in 1995-2017 is greater than 0, the p-value is less than 0.01, which 126 indicates the carbon emission intensity of each province has a significant positive spatial correlation. 127 Secondly, the Moran's I index from 1995 to 2017 generally shows an upward trend, indicating that the 128 accumulation effect of inter-provincial carbon emission intensity tends to strengthen, and the 129 differences among different provinces have gradually widened. 130 Table 2 Moran's I index 131 Year The spatial panel model can be modified by the least-squares regression model. Therefore, this paper 135

(6)
The superiority of the SEM is reflected in the careful consideration of other factors affecting the carbon 141 emission intensity of each province, and the other factors are represented as random error terms. The 142 model is as follows: 143 ln cg = β 0 + β 1 ln + β 2 ln + β 3 ln pop + β 4 ln ly + β 5 ln es + β 6 ln is + (7) β 7 ln rd + β 8 ln fd + μ + λ + ε, ε = δWε + φ The explained variable itself may have spatial correlation, and the explanatory variable and error term 144 may also have spatial characteristics. The SDM can reflect the spatial correlation from different sources 145 and be modified into SLM and SEM by setting different coefficients. Based on this, this article chooses 146 a more general SDM for analysis, and the SDM is as follows: 147 ln cg = β 0 + ρW ln cg + β 1 ln + β 2 ln + β 3 ln pop + β 4 ln ly + β 5 ln + β 6 ln is + β 7 ln rd + β 8 ln fd + θ 1 ln + θ 2 ln + θ 3 ln pop + θ 4 ln ly + θ 5 ln + θ 6 ln + θ 7 ln + θ 8 ln In addition, there is a path-dependent characteristic of carbon emission changes from the time 148 dimension, that is, the time lag effect. There may also be a two-way causal relationship between carbon 149 emissions and factors such as economic growth and technological progress, resulting in endogenous 150 problems (Shuai et al., 2011). Therefore, the lag phase of the carbon intensity variable was introduced 151 into the standard static SDM. The dynamic SDM is as follows. 152 ln cg = β 0 + ln −1 + ρW ln cg + β 1 ln + β 2 ln + β 3 ln pop + β 4 ln ly + β 5 ln + β 6 ln is + β 7 ln rd + β 8 ln fd + θ 1 ln + θ 2 ln + θ 3 ln pop + θ 4 ln ly + θ 5 ln + θ 6 ln + θ 7 ln + θ 8 ln The weight matrix is divided into 3 categories, namely 0-1 weight matrix, geographic distance weight 153 matrix, and economic distance weight matrix. The geographical distance weight matrix is the most 154 common, represented as the reciprocal of the geographical distance of each province, and the formula 155 is as follows. 156

Intermediary effect 157
The mediating effect refers to the indirect effect of explanatory variables on the explained variables 158 through intermediate variables (Mackinnon et al., 2000), tested by the widely used stepwise method 159 (Baron and Kenny, 1999). The testing process is based on the following two conditions: the 160 explanatory variable significantly affects the explained variable, and subsequent variables in the causal 161 chain will be significantly affected by the previous variable. Specifically, the explanatory variable (X) 162 has an indirect effect on the explained variable (Y) through the intermediate variable (M). Thus, the 163 conduction process is as follows. 164 In order to better construct the spatial model, the Lagrange multiplier test, likelihood ratio test, and 168

Result and Discussion
Hausman test can empirically examine the scientific nature of the spatial panel Durbin model. The 169 p-value of the houseman test is close to 0. At the same time, this article pays more attention to the 170 changes of specific individuals within the region. Therefore, both test results and theory support the use 171 of the fixed-effects model (Baltagi, 2008). 172 Lagrange multiplier test and likelihood ratio test provide guidance for the choice of spatial models 173 (Anselin and Florax, 1995; Burridge and Fingleton, 2010). According to Table 3, the LM-Error is more 174 significant than LM-LAG, and R-LM Error is more significant than R-LM Lag, so selecting SEM is 175 more appropriate than the SLM. According to

Spatial and temporal characteristics analysis of carbon emission intensity 185
ArcGIS can be used to visualize China's carbon emission intensity and economic agglomeration level 186 in 1997-2017 (Fig. 2). In the past ten years, China's total carbon emission intensity has shown a 187 downward trend. Divided by geographical location, carbon emission intensity decreases from west to 188 east and from north to south. The changing trend of the level of economic agglomeration is opposite to 189 the intensity of carbon emissions, and the degree of economic agglomeration in the eastern coastal 190 areas has always been maintained at a relatively high level.  For the convenience of comparison, the estimated results of the static SDM with fixed spatial effects, 198 fixed time effects, and two-way fixed effects of time and space are reported in Table 5. In order to 199 avoid the endogenous problem caused by economic agglomeration, the dynamic SDM with two-way 200 fixed effects in time and space introduces a lagging one-phase variable of carbon emission intensity. 201 Therefore, the results of this model are more reliable and will be discussed later.

202
The results show that the spatial lag coefficient of carbon emission intensity is significantly positive 203 (Table. 5), indicating that carbon emission intensity has a strong path dependence and a "snowball" 204 effect in the time dimension. In order to achieve the goal of carbon neutrality, China's carbon emission 205 reduction work is both urgent and arduous. The spatial lag coefficient of economic agglomeration is 206 significantly negative, indicating that economic agglomeration in neighboring provinces has a 207 depressing effect on local carbon emission intensity. With the construction of urban agglomerations, the 208 economic ties between neighboring regions have been continuously strengthened. Related industries 209 and enterprises form a specialized division of labor within the entire urban agglomeration. When a 210 central area appears in the urban agglomeration, the central area will continue to attract emerging 211 industries to enter, thereby weakening the attractiveness of the surrounding areas due to the siphon 212 effect. As a result, there is a negative correlation between the degree of economic agglomeration and 213 the intensity of carbon dioxide emissions between regions.

214
The coefficients of the primary, secondary and tertiary terms of economic agglomeration have all 215 passed the 1% significance level test, and there is a significant inverted-N relationship between 216 economic agglomeration and carbon emission intensity. In the early stage of economic development, 217 industrial gatherings had a restraining effect on carbon emission intensity. In the early days of China's 218 reform and opening up, urbanization and industrialization were both in their infancy. When the number 219 and scale of enterprises are limited, the scale effect promotes production efficiency, and the 220 infrastructure can be shared. As a result, the speed of economic agglomeration greatly exceeds the 221 intensity of carbon emissions. With the promotion of China's urbanization process, economic 222 agglomeration plays a role in promoting carbon emission intensity. During this period, the expansion of 223 enterprise production scale leads to an increase in the demand for production factors, so the carbon 224 emission intensity in the production stage increases. In the period of rapid economic development in 225 China at the beginning of the 21st century, on the one hand, the Yangtze River Delta and the Pearl 226 River Delta have become the world's foundries. 227 On the other hand, environmental regulatory policies, land protection policies, and the promotion of 228 clean technologies lag behind the growth of economic agglomeration. They are leading to economic 229 development and increasing the intensity of carbon emissions. In the end, as the strength of enterprises 230 increases, the division of specialization is strengthened, and environmental regulations and policies are 231 improved-the increase in the cost of environmental pollution forces enterprises to reduce carbon 232 emissions. 233 "***", "**" and "*" indicate significance at the 1%, 5%, and 10% levels, respectively, but not significant if not 235 marked; p-value in parentheses 236

Spatial effects of energy intensity 237
The spatial lag coefficient of energy intensity is significantly positive at the 1% level (Table 5). On the 238 one hand, the energy intensity of neighboring provinces positively impacts the province's carbon 239 emission intensity. On the other hand, it also confirms the existence of regional economic competition 240 and imitation effects. When neighboring cities develop by supporting high-pollution and high-emission 241 industries, neighboring cities will also choose to imitate. With the development of the industrial chain 242 of urban agglomerations, energy consumption in adjacent areas tends to be similar, and carbon 243 emission intensity also positively correlates. 244 The signs of the primary and secondary coefficients of energy intensity are negative and positive, 245 respectively, and both are significant at the 1% level (Table 5), indicating a typical inverted U-shaped 246 curve relationship between energy and carbon emission intensity. In the early stage of economic 247 development, the expansion of production triggered by economic agglomeration promoted an 248 accelerated increase in energy consumption and carbon emission intensity. However, with the 249 implementation of energy-saving and emission reduction policies and technological progress, energy 250 use efficiency has been improved, energy consumption structure has been optimized, and clean 251 technology has been popularized. Ultimately, the carbon emission intensity will show a downward 252 trend again. 253

Effects of the control variable 254
From the perspective of control variables, per capita income and its quadratic coefficient are 255 significantly negative and significantly positive, respectively, indicating a U-shaped curve relationship 256 between per capita income and carbon emission intensity. The coefficient of energy structure is 257 significantly positive, confirming that China's production model of relying on coal resources limits the 258 decline in carbon emission intensity. The coefficient of industrial structure is significantly positive, 259 indicating that excessive dependence on the secondary industry is not conducive to reducing carbon 260 emission intensity. The coefficient of technological progress is significantly positive, indicating that the 261 improvement in production efficiency and the development of clean technology caused by 262 technological progress has played a role in promoting energy conservation and emission reduction. The 263 population size is significantly positive, indicating that population agglomeration will increase carbon 264 emissions in the region. Finally, the coefficient of openness to the outside world is significantly 265 positive, verifying the pollution refuge hypothesis, that is, China has attracted foreign investment in 266 high-carbon emission industries. 267

Energy intensity intermediary effect 268
The stepwise method is a suitable method to test whether energy intensity acts as an intermediary 269 variable for economic agglomeration and carbon emission intensity. At the same time, whether there is 270 an energy-saving effect in economic agglomeration can also be verified.

271
According to Table 6, there is an inverted N-shaped relationship between economic agglomeration and 272 carbon intensity, a U-shaped curve relationship between economic agglomeration and energy intensity, 273 and an inverted U-shaped curve relationship between energy intensity and carbon intensity. Based on 274 the empirical results of the intermediary effect model, it is verified that energy intensity is an 275 intermediary variable between economic agglomeration and carbon emission intensity, and it also 276 validates the energy-saving and emission-reduction effects of economic agglomeration. In addition, the 277 results of models 5 and 7 also confirmed the robustness of the model 4 dynamic space Doberman 278 model. 279 Table 6 Intermediary  "***", "**" and "*" indicate significance at the 1%, 5%, and 10% levels, respectively, but not significant if not 281 marked; p-value in parentheses 282

Spatial heterogeneity analysis
283 Table 5 shows the impact of economic agglomeration in the three regions on carbon emission intensity. 284 Due to the time lag effect of carbon emission intensity, it is more appropriate to use the dynamic spatial 285 Doberman model. The internal economic agglomeration and carbon emission intensity of the three 286 regions present an "inverted N" relationship, and the energy intensity and carbon emission intensity 287 present an inverted U-shaped relationship, which is consistent with the overall national trend. In 288 addition, economic agglomeration and energy intensity have a significant spatial spillover effect on 289 carbon emission intensity. Therefore, the increase in the level of economic agglomeration in the region 290 will reduce the carbon emission intensity of the surrounding areas. In contrast, the increase in the 291 region's energy intensity will promote the increase of the carbon emission intensity of the surrounding 292 areas.

293
Comparing the differences between the three regions, economic agglomeration in the western region 294 has the most apparent inhibitory effect on carbon emission intensity, and energy intensity also has the 295 most significant promotion effect on carbon emission intensity. Due to geographical location and 296 historical factors, the development of western China has always lagged behind other regions. Therefore, 297 when policies such as the development of the Chengdu-Chongqing urban agglomeration and the New 298 Silk Road are being promoted, attention should also be paid to energy conservation and emission 299 reduction in the western region to China's environmental improvement. 300 Table 7 Spatial heterogeneity estimation results 301

315
With the steady progress of China's new urbanization, urban agglomeration economy, and the "two 316 belts and one road" regional development strategies, "group-style" development and industrial 317 agglomeration have become the driving force of China's future economic growth. However, in the face 318 of the bright future of carbon neutrality in 2060 and the reality of high carbon emissions in various 319 provinces, achieving win-win results with energy conservation and emission reduction has become an 320 urgent problem to be solved. This paper uses data from 30 provinces in China from 1995 to 2017, 321 innovatively considers the time lag and spatial spillover effects of carbon emission intensity, and 322 introduces energy intensity as an intermediary variable. It also analyzes the internal mechanism of 323 energy conservation and emission reduction of economic agglomeration and provides policy 324 recommendations for China's economic green transformation development and regional development 325 strategies.

326
According to the research results, in the time dimension, carbon emission intensity has path-dependent 327 characteristics, and the "snowball" effect is prominent. In the spatial dimension, both economic 328 agglomeration and energy intensity show spatial solid spillover effects. The carbon emission intensity 329 of this region is highly susceptible to the influence of the economic development model and energy 330 consumption intensity of the surrounding areas. Second, economic agglomeration can play a role in 331 energy conservation and emission reduction through positive externalities such as technology spillovers, 332 facility sharing, centralized supervision, and specialized division of labor. Finally, with energy intensity 333 as an intermediary variable, there are direct and indirect effects on the mechanism of economic 334 agglomeration on carbon emission intensity. 335 Based on the research conclusions, this article puts forward the following policy recommendations: 336 First, urban agglomerations have become the main spatial form of new urbanization, and the trend of 337 regional economic integration remains unchanged. Therefore, China should pay attention to the 338 economic development of the central and western regions and give full play to the optimal energy 339 conservation and emission reduction of economic agglomeration effect. Second, in policy formulation, 340 energy conservation and emission reduction policies should be coordinated with each other, and the 341 achievement of emission reduction targets needs to be consistent with energy conservation targets. 342 Finally, due to the significant spatial spillover effects of economic agglomeration and energy intensity, 343 it is necessary to reach a coordinated governance mechanism for energy conservation and emission 344 reduction policies between regions to promote the regional linkage mechanism of China

358
The authors declare that they have no conflict of interest. 359

360
No funding was received for conducting this study. 361 All authors certify that they have no affiliations with or involvement in any organization or entity with 362 any financial interest or non-financial interest in the subject matter or materials discussed in this 363 manuscript. 364

365
Some or all data, models, or code generated or used during the study are available from the 366 corresponding author by request. 367