The Impact of Polycentric Structure on CO₂ Emissions: Evidence from China

Abstract

Driven by the 2030 Agenda for Sustainable Development, the importance of sustainable urbanization has taken center stage. In this study, we investigate the impact of polycentric structures on CO₂ emissions using data from 279 Chinese cities and employing two-way fixed effects complemented by instrumental variables. Our findings indicate that polycentric structures effectively alleviate CO₂ emissions. We identify two key pathways through which polycentric structures contribute to CO₂ reduction: promoting green technology and curbing energy consumption. Additionally, we discover that these relationships are influenced by market integration levels and resource dependency. This research offers valuable insights into the future development of sustainable urban spatial structures, paving the way for more eco-friendly cities around the globe.

1. Introduction

The 2030 Agenda for Sustainable Development offers a visionary roadmap for a sustainable future. There is global consensus on coordinating spatial expansion, economic growth, and the environment to achieve sustainable development [1,2]. Sustainable Development Goal 11 (SDG 11) aim to create inclusive, safe, resilient, and sustainable cities and communities for all. However, the increasing threat of climate change looms over our living environments. The most recent progress report on SDG 13, which focuses on climate action, highlights the urgent need to reduce CO₂ emissions, as they reached a historic high in 2021 [3]. In countries like China, rapid urbanization and the concentration of resources in major cities exacerbate CO₂ emissions. Consequently, finding a balance between resource integration and sustainable development has become a critical challenge in urban planning.

Research has shown that adopting a polycentric structure can be beneficial in terms of resource integration and optimizing efficiency [4]. Thus, clarifying the impact of polycentric structure on CO₂ emissions can help direct the practice of sustainable development. However, the relationship between polycentric structure and CO₂ emissions has not reached consensus. Some studies prove that the polycentric structure takes advantage of the integration of resources and optimizes the efficiency of factors [5]. Thus, it can alleviate CO₂ emissions [4,6]. It’s important to note that some studies have raised concerns about polycentricity, suggesting that it may lead to higher household emissions, increased industrial manufacturing [7], and greater household heating demands [8], all of which could contribute to higher CO₂ emissions. While the other studies demonstrate a non-linear relationship between polycentric structure and CO₂ emissions [9,10]. This highlights the need for further exploration and discussion to better understand the intricacies of this relationship.

The inconsistencies in the relationship between polycentricity and CO₂ emissions can be explained by four reasons: First, this relationship stems from a balance among diverse influence pathways. Previous studies have examined pathways such as reduced commuting duration, traffic congestion, household carbon emissions, and industrial emissions [7,11]. Focusing on different pathways could lead to inconsistent findings regarding the connection between polycentric structures and CO₂ emissions. Second, the sample areas vary from study to study. For example, Burgalassi and Luzzati [7] found that polycentric structure aggravates CO₂ emissions in Italy, while Sun et al. [6] and Sha et al. [11] discovered a positive association between polycentricity and CO₂ mitigation in China. One plausible reason is that the impact of polycentric structures on CO₂ emissions may be influenced by factors such as regional resource dependence and the level of market integration. Third, even within the same country, diverse scales of study regions may lead to a different association between polycentric structures and CO₂ emissions. The evidence from Chinese cities showed a negative association between polycentric structure and CO₂ [11], while the evidence from Chinese provinces demonstrated a U-shaped relationship [9]. Fourth, the measurement of core variables and the robustness of methods can also influence the results. The influence of the morphological polycentric index differs from that of the functional dimensions of polycentric structures. Additionally, conducting multiple robust tests is recommended to mitigate endogeneity and determine more accurate associations between spatial structures and CO₂ emissions [6].

This study evaluates the impact of morphological polycentricity on CO₂ emissions in China using a panel dataset from 279 cities spanning from 1999 to 2017. Then we propose two plausible pathways, green technology and energy consumption, and discussed the heterogeneous effects of market integration and resource dependence. Notably, we introduce an innovative instrument variable (IV) in conjunction with a widely used IV to mitigate the endogeneity, exclude the interferences of policies, adopt propensity score matching (PSM) to avoid selection bias, alter fixed effects, clustered levels, and core variable measurements to ensure the robustness of our findings. The contributions of this study are as follows: First, we provide robust evidence of the impact of polycentric structures on CO₂ emissions in China, accounting for potential biases from endogeneity and policy-related factors. Second, we discuss the plausible mechanisms that influence this relationship by examining the mediating effects of green technology and energy consumption, thus enhancing our understanding of the spatial structure and CO₂ emissions framework. Third, we highlight the heterogeneous effects of market integration and resource dependence, which partly explains the inconsistent results among previous studies.

The rest of the study is organized as follows: Section 2 reveals the data and methods adopted in this study. Section 3 presents the impact of polycentric structure on CO₂ emissions and shows the robustness, mechanisms, and heterogeneity of this relationship. Section 4 analyzes the results above. Section 5 summarizes the conclusion and its policy implications. The conceptual framework is represented in Figure 1.

2. Materials and Methods

2.1. Data and Variable

2.1.1. Dependent Variables

We adopted the logarithm of regional CO₂ emissions as the dependent variable. The data on CO₂ emissions were generated by Chen et al. [12], who used a particle swarm optimization-back propagation algorithm to unify the scale of DMSP/OLS and NPP/VIIRS satellite imagery at the county level. We carried out our study using the data of the CO₂ emissions between 1999 and 2017.

2.1.2. Independent Variable

In this study, we focused on morphological polycentricity, which is a balanced distribution of the importance of the cities within a given region. Generally, it can be quantified by employment, population size, or economic output [6]. We used the Prolonged Artificial Nighttime-Light Dataset of China generated by Zhang et al. [13] to measure the economic activity of cities [10]. Based on this dataset, we measured the polycentricity as the Pareto exponent, which is calculated by the slope of the size and the ranks of each county within a city [14]:

$$\ln \left(Ran{k}_j\right)=A-QlnLigh{t}_j$$

(1)

where Q refers to the level of polycentricity. The higher the Q-value, the more polycentric and compact a city’s spatial structure is. $Ran{k}_j$ is the rank of economic activity of county j in descending order within the given city. $lnsiz{e}_j$ is the economic activity level of county j.

2.1.3. Control Variables

Following previous studies on CO₂ emissions, we controlled secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population to exclude their influence on CO₂ emissions [15,16].

Table 1. Definitions and operations of variables.

Variable	Definition
Dependent variable
Carbon	The logarithm of regional CO2 emissions.
lndependent variables
Q	The Pareto exponent.
Mediating variable
Gti	The logarithm of the annual number of green patent applications plus one.
Energy	The logarithm of electricity consumption.
Control variables
lngy	The share of the secondary sector in GDP.
lnFDI	The logarithm of the ratio of the region’s actual use of foreign capital to GDP in that year.
lnpeople	The logarithm of the ratio of the total population to the local land area at the end of the year.
lnfis	The logarithm of the ratio of regional general budget fiscal expenditure to GDP.
lnsci	The logarithm of the ratio of regional science and technology expenditure to GDP.
lnurba	The logarithm of the ratio of the urban employed population to the year-end household population.
Other variables
Hlhl	The ratio of the river density to the exchange rate.
Unemploy	The ratio of unemployment to the population.
LCC	If it is the pilot city of Low-carbon city policy, this variable = 1, 0 otherwise.
ETS	If it is the pilot city of the emissions trading scheme, this variable = 1, 0 otherwise.
Rep_carbon	The CO2 emissions proposed by of Wu et al. [17]
PCE	CO2 emissions per capita.
Lngreencov	The logarithm of canopy coverage.
Lnlpg	The logarithm of household consumption of liquefied petroleum gas.
Lnws	The logarithm of average wind speed.
Lnhum	The logarithm of average relative humidity.
Lntem	The logarithm of average temperature.

defined all the variables used in this study.

2.2. Data Summary

Descriptive statistics for each variable were reported in

Table 2. Descriptive statistics of variables.

Variables	Obs.	Mean	S.D.	Min	Max
Carbon	5278	2.743	0.858	0.0881	5.441
Q	5269	0.924	0.540	0.252	5.498
lngy	5226	3.868	0.282	2.086	4.525
lnFDI	4672	−6.201	1.386	−15.250	−1.816
lnpeople	5295	−2.717	0.973	−7.624	0.340
lnfis	5290	−2.094	0.504	−4.221	0.994
lnsci	5279	11.370	1.270	5.727	15.430
lnurba	5240	−1.809	0.665	−3.358	1.249
Hlhl	5330	0.291	0.209	0.00577	2.278
Unemploy	5278	3.884	0.697	−0.486	7.051
LCC	5330	0.139	0.346	0	1
ETS	5330	0.0356	0.185	0	1
Rep_carbon	5071	7.841	0.569	6.047	11.51
PCE	5275	−3.104	0.765	−5.931	−0.270
Win_carbon	5278	2.742	0.844	0.739	4.652
Gti	4968	3.225	1.683	0.693	8.897
Energy	2789	13.070	1.221	4.625	16.54
Lngreencov	5243	3.488	0.459	−1.715	5.957
Lnlpg	5125	8.820	1.510	1.386	13.809
Lnws	5330	0.735	0.266	−0.302	1.582
Lnhum	5330	4.227	0.143	3.590	4.481
Lntem	5323	2.588	0.482	−2.036	3.274

. The overall CO₂ emissions had a value of 2.743 on average. The average polycentricity was 0.924. Generally, if the q-value is greater than 1, the spatial structure of the city is polycentric. If it is less than 1, the spatial structure is relatively monocentric.

2.3. Methodology

Accounting for unit-variant and time-variant variables, we adopted a two-way fixed effects model to control for both unit- and time-fixed effects. We constructed the following model:

$$Carbo{n}_{i,t}={\alpha }_0+ {\alpha }_1{Q}_{i,t}+\sum {\tau }_j\ast X_{j,i,t}{+\mu }_i+{\phi }_t+{\epsilon }_{i,t}$$

(2)

where the subscript i represents different cities, t represents time variables, $Carbo{n}_{i,t}$ represents the natural logarithm of regional CO₂ emissions, $Q_{i,t}$ represents the polycentricity of city i, $X_{j,i,t}$ is the matrix-vector of the control variables, ${\mu }_i$ represents a city-specific effect to control for unobserved heterogeneity, ${\phi }_t$ represents the control for the year-fixed effect, and ${\epsilon }_{i,t}$ is the stochastic disturbance term.

To mitigate concerns about endogeneity and figure out the causality between polycentric spatial structures and CO₂ emissions, we constructed two IVs that were correlated with polycentric spatial structures but met the exogeneity condition of CO₂. Specifically, we followed Chen et al. [10] and used the river density/exchange rate (i.e., Hlhl) as the first instrument for polycentricity [10]. Proximity to water sources could affect the spatial structure of cities [18]. In the meantime, it is not related to CO₂ emissions. Thus, river density is a proper IV for polycentricity.

We then proposed an innovative instrument, namely, the proportion of unemployment (i.e., unemployment as a percentage of the population). It has been proven that there are interdependencies between labor market behavior and the spatial structure of cities [19]. A large proportion of unemployed people indicates that the city has developed excessively or poorly, which leads to a lack of impetus for further expansion of the city, which has a negative impact on polycentricity. Meanwhile, the proportion of unemployment cannot affect CO₂.

Finally, the ratio of river density to exchange rate and the proportion of unemployment were IVs for polycentric spatial structure. The first stage of two-stage least squares (2SLS) was as follows:

$$Q_{i,t}={\beta }_0+{\beta }_1{Hlhl}_{i,t}+{\beta }_2{\mathit{Unemploy}}_{i,t}+\sum {\tau }_{\mathrm{j}}\ast X_{\mathrm{j},i,t}{+\mu }_i+{\phi }_t+{\epsilon }_{i,t}$$

(3)

where ${\mathit{Hlhl}}_{i,t}$ and ${\mathit{Unemploy}}_{i,t}$ represent the ratio of river density to the exchange rate and the proportion of unemployment, respectively.

The second stage of 2SLS was as follows:

$$Carbo{n}_{i,t}={\delta }_0+{\delta }_1\widehat{Q_{i,t}}+\sum {\tau }_{\mathrm{j}}\ast X_{\mathrm{j},i,t}{+\mu }_i+{\phi }_t+{\epsilon }_{i,t}$$

(4)

where $\widehat{Q_{i,t}}$ is the first-stage predicted value for the polycentricity (i.e., $Q_{i,t}$).

Previous studies have verified the mediating role of transportation, household carbon emissions, and industrial emissions in the relationship between polycentric structures and CO₂ emissions [6,11]. We proposed two innovative pathways to examine how polycentric structures affect CO₂ emissions. We hypothesized that green technology and energy consumption might serve as potential mechanisms through which polycentric structures influence CO₂ emissions. To test these hypotheses, we measured green technology as the logarithm of the annual number of green patent applications plus one assessing its mediating effect on the relationship. For energy consumption, we used the logarithm of electricity consumption to evaluate its potential mediating role in the relationship between polycentricity and CO₂ emissions. With these variables in place, we constructed the following model to analyze the impact of polycentric structures on CO₂ emissions through these proposed pathways:

$$M_{i,t}={\gamma }_0+{\gamma }_1{Q}_{i,t}+\sum {\tau }_j\ast X_{j,i,t}{+\mu }_i+{\phi }_t+{\epsilon }_{i,t}$$

(5)

where $M_{i,t}$ represents mediators (i.e., green technology and energy consumption).

3. Results

3.1. Effect of the Polycentric Spatial Structures on CO₂ Emissions

Based on Equation (1), Columns (1)–(3) of

Table 3. The impact of polycentric spatial structures on CO₂ emissions.

	(1)	(2)	(3)	(4)	(5)
	FE	FE_DK	FE_BS	First Stage	Second Stage
Variables	Carbon	Carbon	Carbon	Q	Carbon
Q	−0.050 ***	−0.050 ***	−0.050 **		−0.772 ***
	(−2.61)	(−3.32)	(−2.48)		(−4.06)
Hlhl				5.233 ***
				(3.57)
Unemploy				−0.029 ***
				(−4.31)
Control	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	4513	4513	4513	4496	4496
R-squared	0.947	0.947	0.947		0.882
K-P LM statistic					24.219 ***
K-P Wald F statistic					12.663
Hansen J statistic					0.219

Notes: t/z-values are in parentheses; **, and *** represent the 5%, and 1% significance levels, respectively; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space; the unit of river/exchange rate is %.

reported the results with a two-way fixed effect with robust standard errors, Driscoll-Kraay standard errors [20], and 1000 bootstrap replications, respectively. The results demonstrated a significant negative effect of polycentricity on CO₂ emissions, which implied that polycentric spatial structures helped reduce CO₂ emissions in China.

3.2. Robustness Checks

3.2.1. Endogeneity

The results based on the 2SLS estimator were presented in Columns (4) and (5) of Table 3. The estimation results of the first stage demonstrated that the associations between IVs and polycentricity were significant at a 1% level, and the F-statistic is 12.663, surpassing the threshold for a strong instrument. The K-P LM statistic was significant at a 1% level, indicating the IVs were related to the polycentric spatial structure. And the Hansen J-test indicated that there were no over-identification problems and the IVs were strictly exogenous.

The results suggested that polycentric spatial structure supports CO₂ emission reduction after considering endogeneity. Compared with the coefficients in Columns (1)–(3), the coefficients estimated by 2SLS were greater than those estimated by the two-way fixed effect. One plausible reason was that there might be reverse causality between polycentric spatial structure and CO₂ emission because air pollution could also affect the labor market, which would influence the spatial structure of cities [21]. The 2SLS approach helped mitigate this potential source of bias.

3.2.2. Alter Fixed Effects and Clustered Levels

In the baseline results, we controlled both individual-fixed effects and year-fixed effects. To avoid the influence of unmeasured factors, we controlled for fixed effects in the dimension of the province. Furthermore, we adopted interactive clustering that took different clustering levels, such as city, province, and the interaction term of these two levels, into account. The results for altering fixed effects and clustering levels were shown in Figure 2. The blue line showed the 95% confidence interval, the orange point presented the coefficient, and the red dotted line indicated the 0 value. The results indicated that polycentric spatial structure had a significantly negative impact on CO₂ emissions. The fixed effect of province dimension and diverse cluster levels did not influence this relationship; namely, the benchmark results were robust after considering these unmeasured factors.

3.2.3. Exclude the Interference of Related Policies

CO₂ emissions highly interfered with relevant policies in China. For example, the low-carbon city (LCC) policy emphasized CO₂ emissions along with economic development [22]. The effectiveness of LCC in CO₂ emissions abatement had been verified in pilot cities [23]. Furthermore, the emissions trading scheme (ETS), as a typical market-oriented environmental instrument, was also conducive to reducing CO₂ emissions [24]. In this regard, the interference of LCC and ETS might confound the relationship between polycentricity and CO₂ emissions.

Thus, we controlled these two policies based on the benchmark results to mitigate their interference. The results were presented in Columns (1) and (2) of

Table 4. Robustness check: the impact of polycentric spatial structures on CO₂ emissions.

	(1)	(2)	(3)	(4)	(5)
Variables	Carbon	Carbon	Rep_Carbon	PCE	Win_Carbon
Q	−0.045 **	−0.045 **	−0.018 **	−0.128 ***	−0.055 ***
	(−2.33)	(−2.42)	(−2.52)	(−2.67)	(−2.92)
LCC	−0.067 ***
	(−4.26)
ETS		−0.119 ***
		(−5.83)
Control	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	4513	4513	4313	4513	4513
R-squared	0.948	0.949	0.850	0.915	0.941

Notes: t-values are in parentheses; **, and *** represent the 5%, and 1% significance levels, respectively; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space.

. After controlling the relevant policies (i.e., LCC and ETS), respectively, the coefficients of polycentricity are still significantly negative. It indicated that polycentric spatial structures helped reduce CO₂ emissions; namely, the benchmark results are robust.

3.2.4. Alter the Measurement of CO₂ Emissions

The impact of polycentricity on CO₂ emissions might also be influenced by the measurement of CO₂ emissions. To guarantee the robustness of benchmark results, we measured the CO₂ emissions by Wu et al. [17] and CO₂ emissions per capita. The estimated results were shown in Columns (3) and (4) of Table 4, respectively. The impact of polycentric spatial structure helped CO₂ abatement, indicating that altering the measurement of CO₂ emissions did not influence the benchmark results.

Furthermore, the results of a regression could be sensitive to outliers, while the CO₂ emissions varied widely among 4513 city-year observations (minimum = 0.0881, maximum = 5.441). Thus, we tailored the data at the 1% level. The results were presented in Column (5) of Table 4. After excluding the outliers in CO₂ emissions, the polycentricity still had a significant and negative impact on CO₂ emissions, indicating that the benchmark results were robust.

3.2.5. PSM

One might argue that the CO₂ abatement was not achieved by polycentric spatial structure but by the selection bias between polycentric structure and monocentric structure. To mitigate such selection bias, we adopted propensity scores to match cities in the polycentric structure with those in the monocentric structure and make them comparable. The propensity score is estimated using a logit regression model. Columns (1)–(3) in

Table 5. Robustness check: the impact of polycentric spatial structures on CO₂ emissions using PSM.

	(1)	(2)	(3)
Variables	Carbon	Carbon	Carbon
Q	−0.0558 ***	−0.0508 ***	−0.0508 ***
	(−3.15)	(−2.64)	(−2.64)
Control	YES	YES	YES
City FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	3771	4508	4508
R-squared	0.947	0.947	0.947

Notes: t-values are in parentheses; *** represent the 1% significance levels, respectively; population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population are adopted to match cities in the polycentric structure with those in the monocentric structure; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space.

reported the results using nearest neighbor matching (1:4), caliper matching, and kernel matching, respectively. The results of balance tests were reported in Appendix A 

Table A1. Balance Test of the Nearest Neighbor Matching (1:4).

Variables	Unmatched	Mean		Bias (%)	Reduce Bias (%)	t-Test
	Matched	Treated	Control			T	P > T
lnpeople	U	−2.6799	−2.7279	5.0		1.81	0.071
	M	−2.6798	−2.7092	3.1	38.8	1.13	0.259
lnfis	U	−2.0714	−2.1343	12.7		4.57	0.000
	M	−2.0712	−2.0502	−4.2	66.7	−1.52	0.127
lnsci	U	11.538	11.195	27.3		9.80	0.000
	M	11.539	11.577	−3.0	88.9	−1.11	0.266
lnurba	U	−1.8564	−1.7648	−13.9		−5.00	0.000
	M	−1.8568	−1.8376	−2.9	79.0	−1.06	0.289

Notes: In the second column, U refers to the sample before employing the PSM, while M denotes the matched sample after applying the PSM.

and Figure A1,

Table A2. Balance Test of Calliper Matching.

Variables	Unmatched	Mean		Bias (%)	Reduce Bias (%)	t-Test
	Matched	Treated	Control			T	P > T
lnpeople	U	−2.6799	−2.7279	5.0		1.81	0.071
	M	−2.6800	−2.7130	3.5	31.1	1.27	0.204
lnfis	U	−2.0714	−2.1343	12.7		4.57	0.000
	M	−2.0710	−2.0606	−2.1	83.4	−0.76	0.445
lnsci	U	11.538	11.195	27.3		9.80	0.000
	M	11.538	11.554	−1.3	95.4	−0.46	0.643
lnurba	U	−1.8564	−1.7648	−13.9		−5.00	0.000
	M	−1.8565	−1.8474	−1.4	90.2	−0.50	0.618

Notes: In the second column, U refers to the sample before employing the PSM, while M denotes the matched sample after applying the PSM.

and Figure A2, and

Table A3. Balance Test of Kernel Matching.

Variables	Unmatched	Mean		Bias (%)	Reduce Bias (%)	t-Test
	Matched	Treated	Control			T	P > T
lnpeople	U	−2.6799	−2.7279	5.0		1.81	0.071
	M	−2.6798	−2.7059	2.7	45.6	1.01	0.315
lnfis	U	−2.0714	−2.1343	12.7		4.57	0.000
	M	−2.0712	−2.0654	−1.2	90.8	−0.42	0.672
lnsci	U	11.538	11.195	27.3		9.80	0.000
	M	11.539	11.531	0.6	97.8	0.22	0.826
lnurba	U	−1.8564	−1.7648	−13.9		−5.00	0.000
	M	−1.8568	−1.8391	−2.7	80.7	−0.98	0.328

Notes: In the second column, U refers to the sample before employing the PSM, while M denotes the matched sample after applying the PSM.

and Figure A3, respectively. The results of balance tests demonstrated that after matching, there was no significant bias between cities with polycentric and monocentric structures, indicating that PSM helped mitigate selection bias. After matching, the impact of polycentric spatial structures on CO₂ emissions was still significantly negative, implying a positive role for polycentricity in CO₂ reduction.

3.2.6. Exclude Other Confounders

Despite accounting for various confounding factors that could influence the relationship between polycentric structures and CO₂ emissions in our baseline results, there were still some potential confounders that we have not considered. Factors such as canopy coverage [25], energy consumption [26], wind speed [27], humidity [28], and temperature [29] may affect carbon emissions. To minimize the impact of these confounders, we included them as controls in our baseline results. Columns (1)–(5) in

Table 6. Robustness check: The impact of polycentric spatial structures on CO₂ emissions.

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Carbon	Carbon	Carbon	Carbon	Carbon	Carbon
Q	−0.052 ***	−0.051 **	−0.050 ***	−0.050 ***	−0.050 ***	−0.051 ***
	(−2.77)	(−2.55)	(−2.60)	(−2.60)	(−2.59)	(−2.67)
lngreencov	0.030					0.031
	(1.43)					(1.38)
lnlpg		0.003				0.002
		(0.004)				(0.41)
lnws			−0.005			−0.014
			(0.047)			(−0.28)
lnhum				−0.162 *		−0.162 *
				(−1.91)		(−1.81)
lntem					0.022	−0.007
					(0.85)	(−0.30)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	4473	4400	4513	4513	4510	4364
R-squared	0.948	0.947	0.947	0.947	0.947	0.948

Notes: t-values are in parentheses; *, **, and *** represent the 10%, 5%, and 1% significance levels, respectively; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space.

displayed the effect of polycentric spatial structures on CO₂ emissions after controlling for canopy coverage, energy consumption, wind speed, humidity, and temperature, respectively. The results indicated that polycentric structures reduce carbon emissions even after accounting for the confounding influences of these additional factors. Column (6) presented the results after controlling for all potential confounders, and the findings remained robust.

3.3. Possible Mechanisms

The analysis above demonstrated that polycentricity had a notable impact on CO₂ emissions. We also highlighted the mediating role that green technology and energy consumption played in the relationship between polycentricity and CO₂ emissions. By understanding this connection, we could develop a more comprehensive understanding of the factors influencing CO₂ emissions and make more informed decisions on sustainable urban development.

The results demonstrated that polycentric structure had a positive impact on green technology (see Columns (1)–(3) in

Table 7. The impacts of green technology and energy consumption on CO₂ emissions.

	(1)	(2)	(3)	(4)	(5)	(6)
Green Technology				Energy Consumption
Variables	FE	FE_DK	FE_BS	FE	FE_DK	FE_BS
Q	0.173 ***	0.173 ***	0.173 ***	−0.121 **	−0.121	−0.121 **
	(3.10)	(3.55)	(2.85)	(−2.36)	(−1.63)	(−2.21)
Control	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	4286	4286	4286	2375	2375	2375
R-squared	0.877	0.877	0.877	0.654	0.654	0.654

Notes: t-values are in parentheses; **, and *** represent the 5%, and 1% significance levels, respectively; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space.

), while had a negative impact on energy consumption (see Columns (4)–(6) in Table 7). It indicated that polycentric structures promoted green technology and reduced energy consumption, both of which were conducive to CO₂ abatement. In this regard, a polycentric structure could reduce CO₂ emissions by promoting green technology and reducing energy consumption.

3.4. Heterogeneity Analysis

The spatial structures and CO₂ emissions were heterogeneous because of the different features in each city. We emphasized the heterogeneity resulting in the administrative hierarchy and the dependence on resources in this context. First, we grouped the sample into municipalities directly under the central government and other cities. Second, we hypothesized that the impact of polycentricity on CO₂ varied between resource-based cities and others because of the resource curse. The empirical results were presented in

Table 8. Heterogeneity analysis for different features of cities.

	(1)	(2)	(3)	(4)
Variables	PCM	Non-PCM	Resource-Based	Non-Resource-Based
Q	−0.064	−0.052 ***	−0.020	−0.055 ***
	(−0.17)	(−2.66)	(−0.55)	(−2.78)
Control	YES	YES	YES	YES
City FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	507	4006	1693	2820
R-squared	0.920	0.951	0.945	0.949

Notes: t-values are in parentheses; *** represent the 1% significance levels, respectively; the coefficients of the control variables (i.e., the secondary industry as a percentage of GDP, foreign direct investment, population density, the percentage of budget fiscal expenditure and technology expenditure on GDP, and the ratio of the employed population), year and time dummies are not reported to save space.

.

Columns (1) and (2) in Table 8 compared the impact of polycentric structure on the CO₂ emissions of municipalities directly under the central government and the others. The results demonstrated that polycentricity had no significant impact on the CO₂ emissions of municipalities while reduced the CO₂ emissions of other cities. Columns (3) and (4) presented the relationship between polycentric structure and CO₂ emissions in resource-based cities and other cities, respectively. The results indicated that the polycentric structure affected CO₂ emissions neither in municipalities nor resource-based cities while significantly reduced CO₂ emissions in other cities.

4. Discussion

Our primary empirical findings suggested that polycentric structures contributed to CO₂ abatement, aligning with previous studies that found increased CO₂ efficiency [11] and decreased CO₂ concentrations [4]. Monocentric structures, which partly represent resource agglomeration in cities, tend to exacerbate emissions [30]. In contrast, polycentric structures can mitigate agglomeration diseconomies, promote market integration, and optimize energy efficiency [10,31], ultimately leading to CO₂ reduction. Our results build upon existing knowledge by using panel data from Chinese city regions and conducting comprehensive robustness tests.

Green technology and energy consumption had been identified as potential mechanisms linking polycentric structures to CO₂ emissions. On the one hand, polycentric spatial structure is conducive to the green economy, which is inevitable for green technology [32,33]. In the meantime, green technology is widely accepted as being environmentally friendly and helping with CO₂ reduction [34]. On the other hand, with the optimization of the polycentric structure, mega-cities form internal subcenters represented by industrial agglomeration, and hence strengthen the level of specialization. Therefore, in the context of coordinated industrial development, the polycentric structure promotes green technology. Furthermore, the polycentric spatial structure might reduce CO₂ through energy consumption. One plausible reason is that polycentric development could affect energy efficiency [35] and, hence, influence energy consumption. Meanwhile, energy consumption is prone to CO₂ emissions [36].

Additionally, the impact of polycentric structures on CO₂ emissions was influenced by administrative hierarchy and resource dependence. Firstly, polycentric structures had no significant influence on CO₂ emissions in municipalities, potentially due to their relatively higher degree of regional market integration. As market segmentation can improve energy efficiency and decrease CO₂ emissions [31], cities with strong market integration might not benefit as much from polycentric structures [10]. Secondly, the polycentric structures did not affect CO₂ emissions in resource-based cities but reduced emissions in other cities. This can be explained by the detrimental impact of the resource curse on environmental sustainability, as resource-based cities often have a single industry structure, which aggravates CO₂ emissions [37]. Resource misallocation decreases energy efficiency and, hence, accelerates CO₂ emissions [38]. Polycentric structures upgrade industry structures [39] and hence have a relatively larger impact on cities with diverse industry structures. Namely, it reduces CO₂ emissions in cities that are not heavily reliant on resources.

5. Conclusions

Fueled by the urgency of achieving the Sustainable Development Goals (SDGs), addressing the rise in CO₂ emissions has become a pressing concern. This study delved into the role of sustainable urban development in mitigating CO₂ emissions by investigating the impact of polycentric structures on CO₂ emissions in 278 Chinese cities from 1999 to 2017. We employed a two-way fixed effect model to estimate our benchmark results and constructed IVs to address endogeneity concerns. Our findings revealed that polycentric structures contributed to CO₂ emissions reduction by promoting green technology and lowering energy consumption. Notably, the positive impact of polycentric structures on CO₂ emissions was only significant in cities that were not classified as municipalities or resource-based cities.

Our study marginally contributed to the theory. First, we figured out a robust relationship between polycentric structures and CO₂ emissions, accounting for potential biases from policy interferences, selection bias, and core variable measurements. Second, we proposed two innovative pathways that mediated the impact of spatial structures on CO₂ emissions. Third, we considered the heterogeneous effects of administrative hierarchy and resource dependence in this context.

From our findings, we derived practical implications for achieving SDGs. First, polycentric structures offer a viable approach for developing countries to regulate CO₂ emissions. When it comes to China specifically, some cities in China should gradually build polycentric spatial structures. Second, government departments should implement strategies related to green technology and energy consumption to regulate CO₂ emissions and achieve SDGs. Third, urban planning and CO₂ mitigation strategies should be tailored to individual cities, with a focus on establishing more environmental regulations in resource-based cities and municipalities.

In closing, we acknowledge some limitations and avenues for future research. First, our study used county-level data to calculate the morphological polycentric index at the city level, overlooking the functional dimensions of polycentric structures. Future studies should explore these functional dimensions further. Second, our analysis focused on the Pareto index in relation to city size balance, suggesting that future research could examine the impact of spatial balance on CO₂ emissions. Lastly, while our study area is China, future studies could investigate the heterogeneous effects of diverse countries and delve into other aspects of the SDGs to gain a more comprehensive understanding of the topic.