To empirically examine the impact of economic growth targets on CO₂ emissions, the study employs panel data of 285 prefecture-level and above Chinese cities from 2005 to 2018. According to Shahnazi and Shabani (2021), CO₂ would spill over because of inter-regional supply chains, pollution haven hypothesis, and strategic interaction behaviors including competition and imitation, resulting in emissions that are not independent across regions. Therefore, this study employs spatial econometric analysis to investigate the spatial effect of economic growth targets on CO₂ emissions. The spatial Durbin model (SDM) encompasses both the spatial autoregressive (SAR) model and the spatial error model (SEM) (LeSage and Pace, 2009), and should therefore be prioritized in empirical analysis according to a general-to-specific strategy (Mur and Angulo, 2009; Chen et al., 2022). The baseline model is set as follows: Y = W Y ρ + X α + W X θ + μ i + ν t + ε i t (1) where Y and X are the explained and explanatory variables, respectively. WY and WX denote the spatial lag terms of the explained and explanatory variables. ρ indicates the spatial autocorrelation coefficient, and α and θ are spatial regressive coefficients. W denotes a row-normalized spatial weight matrix. μ i and ν t represent space and time effects, respectively; ε i t denotes the error term. According to the existing literature, some factors are introduced as control variables, and the specific model is set as follows: ln ⁡ P C E i t = α 0 + α 1 ⁡ ln ⁡ E G T i t + α 2 ⁡ ln ⁡ E L i t + α 3 ⁡ ln ⁡ I S i t + α 4 ⁡ ln ⁡ E O i t + α 5 ⁡ ln ⁡ T I i t + α 6 ⁡ ln ⁡ P D i t + ρ W ⁡ ln ⁡ P C E i t + θ 1 W ⁡ ln ⁡ E G T i t + θ 2 W ⁡ ln ⁡ E L i t + θ 3 W ⁡ ln ⁡ I S i t + θ 4 W ⁡ ln ⁡ E O i t + θ 5 W n T I i t + θ 6 W n P D i t + μ i + ν t + ε i t (2) where lnPCE indicates CO₂ emissions, lnEGT denotes economic growth targets. α is the linear parameter, ρ denotes the spatial regressive parameter of CO₂ emissions, and θ indicates the spatial spillover parameter of economic growth targets on CO₂ emissions.

To identify the transmission mechanisms through which economic growth targets impact CO₂ emissions, we use the mediation effect proposed by Baron and Kenny (1986). Considering the spatial effect, we refer to Huang et al. (2020) and introduce lag terms for various variables. The model is constructed as follows: ln ⁡ M i t = β 0 + β 1 ⁡ ln ⁡ E G T i t + β 2 ⁡ ln ⁡ E L i t + β 3 ⁡ ln ⁡ I S i t + β 4 ⁡ ln ⁡ E O i t + β 5 ⁡ ln ⁡ T I i t + β 6 ⁡ ln ⁡ P D i t + μ i + ν t + ε i t (3) ln ⁡ P C E i t = γ 0 + γ 1 ⁡ ln ⁡ E G T i t + γ 2 ⁡ ln ⁡ M i t + γ 3 ⁡ ln ⁡ E L i t + γ 4 ⁡ ln ⁡ I S i t + γ 5 ⁡ ln ⁡ E O i t + γ 6 ⁡ ln ⁡ T I i t + γ 7 ⁡ ln ⁡ P D i t + ρ W ⁡ ln ⁡ P C E i t + ϕ 1 W ⁡ ln ⁡ E G T i t + ϕ 2 W ⁡ ln ⁡ M i t + ϕ 3 W ⁡ ln ⁡ E L i t + ϕ 4 W ⁡ ln ⁡ I S i t + ϕ 5 W ⁡ ln ⁡ E O i t + ϕ 6 W n T I i t + ϕ 7 W n P D i t + μ i + ν t + ε i t (4) where M denotes the mediating variables, including fiscal decentralization and environmental decentralization.

The distance in geographical space determines the frequency of economic connection, and the difference in economic development affects the degree of resource competition. Consequently, the correlation of carbon emissions between regions may gradually deteriorate with an increasing geographical distance or a growing development gap (Shao et al., 2020). To highlight this feature, we use a geographical distance weight matrix (W ₁ ) to estimate the spatial effect and an economic distance weight matrix (W ₂ ) for robustness checks. Their construction is as follows: W 1 = { 1 d i j , i f   i ≠ j 0 , i f   i = j (5) W 2 = { 1 | G D P i ¯ − G D P j ¯ | , i f   i ≠ j 0 , i f   i = j (6) where d _ij stands for the great circle distance based on longitude and latitude between city i and city j. G D P i ¯ and G D P j ¯ represent the average GDP of city i and city j over the sample period, respectively. W ₁ and W ₂ are standardized by row so that the sum of all elements in any row is 1.

Since most Chinese cities do not set economic growth targets before 2005 or release the number of employees used to measure environmental decentralization after 2018, and 12 cities, including Lhasa, Haidong, and Danzhou, have many missing values in CO₂ emissions, we conduct the empirical analysis using panel data from 285 prefecture-level and above cities from 2005 to 2018. The data come from the EPS database, CEIC database, statistical yearbooks, and government reports. Missing values are supplemented by interpolation and mean. All variables are used as the natural logarithm to process potential heteroscedasticity. For illustration purposes, the definitions and descriptive statistics of variables are presented in Table 1.

The Moran’s I index is employed to analyze the spatial autocorrelation of explained and explanatory variables, and the results with the geographical distance weight matrix are reported in Table 2. From 2005 to 2018, the Moran’s I indexes of lnPCE and lnEGT are positive and pass the 1% significance test, indicating that there is a positive spatial autocorrelation among regions regarding CO₂ emissions and expected targets. Notably, the Moran’s I statistics of lnPCE present a U-shaped curve, and the Moran’s I statistics of lnEGT show a fluctuating upward trend. This indicates the spatial autocorrelation among regions in both CO₂ emissions and economic growth targets appears to be strengthening. As ecological pollution and environmental degradation become increasingly serious, the central government has introduced environmental protection indicators into the official performance evaluation system to transform the GDP-oriented development model, which results in a stronger strategic interaction between local governments on economic growth and environmental governance.

To identify spatial heterogeneity, one needs to perform spatial autocorrelation tests using the local Moran’s index. Figures 1, 2 are scatter plots of the local Moran’s index for lnPCE and lnEGT under the geographical distance weight matrix in 2005 and 2018, respectively. The results show that most of the samples are situated in the first and third quadrants, suggesting that CO₂ emissions and economic growth targets are characterized by high-high and low-low agglomeration. Accordingly, the local Moran’s index illustrates a clear positive spatial autocorrelation among regions in both CO₂ emissions and economic growth targets, in line with the global Moran’s index.

The spatial autocorrelation test indicates that spatial econometric analysis should be used in this paper, and therefore, a series of tests are conducted to select the optimal model. Drawing on LeSage and Pace (2009), the specific steps are shown in Table 3. First, the LM lag, robust-LM lag, LM error, and robust-LM error tests suggest that spatial econometric analysis is more suitable than ordinary regression. Second, the Wald lag, Wald error, LR lag, and LR error tests reject the null hypothesis that the SDM degenerates to SAR or SEM at the 1% significance level. Third, the Hausman test implies that the fixed effect should be adopted over the random effect, and the R-squared and log-likelihood tests show that the space-fixed effect is optimal. Thus, the SDM with a space-fixed effect is selected for the empirical analysis.

The estimation results using SDM are presented in Column (4) of Table 3, and as a comparison, the results using OLS, SAR, and SEM are also given in Columns (1)–(3). As shown in Column (4), the spatial autocorrelation coefficient ρ is 0.275 and is significant at the 1% level, suggesting that every 1% increase in local CO₂ emissions corresponds to a 0.275% increase in surrounding areas. This finding is in keeping with the results of the spatial autocorrelation test. A city’s carbon emissions are influenced not only by the local environment but also by the spatial effects of various factors, such as technology innovation, government intervention, and environmental regulations in neighboring regions (Wenbin Peng et al., 2021).

The decomposition effect results are given in Table 4. The direct effect indicates that a 1% increase in the local economic growth target causes a 0.165% decrease in local CO₂ emissions. That is, economic growth targets have a negative local effect, which reduces local carbon emissions. As environmental governance is linked to performance appraisal, local governments have expanded the single indicator of maximizing GDP into a hybrid indicator that includes prospective economic goals and constrained environmental goals. Thus, local governments tend to develop the economy more rationally than they used to, such as by promoting technological innovation and optimizing industrial structure, and they also use these results to reduce pollution. Notably, the coefficient of lnEGT in Table 4 is slightly different from that in Table 3. This is due to the feedback effect that results from impacts caused by the local region that then spread to neighboring regions and are acted on again in the local region, and it is numerically equal to the difference between the direct and main effects. For example, the direct and main effects of lnEGT are −0.165 and −0.167, respectively, so the feedback effect is equal to 0.002, which means that it weakens the inhibiting effect of economic growth targets on CO₂ emissions.

The indirect effect indicates that a 1% increase in the local economic growth target may induce a 0.909% increase in CO₂ emissions in neighboring regions. This result implies that economic growth targets have a positive spillover effect. On the one hand, local governments will compete for liquidity resources to develop their own economy, which leads to a siphon effect that lowers energy efficiency in neighboring regions. On the other hand, local governments may force high-polluting and high-emission enterprises to relocate to downstream border areas, thereby realizing pollution transfer. Additionally, the indirect effect is much stronger than the direct effect, indicating that the total effect is significantly positive; that is, economic growth targets severely exacerbate carbon emissions.

As for the control variables, the coefficient of the secondary industry’s GDP ratio is statistically insignificant, as determined by Huang et al. (2020). Although there are a variety of high-pollution sectors in the secondary industry, energy consumption can be reduced by restraining the development of other industrial sectors (Shao et al., 2020). The direct effect of lnEO is significantly positive, suggesting that FDI increases local CO₂ emissions. This is in line with the pollution haven hypothesis. The direct and indirect effects of lnEL are significantly negative and positive, respectively, which means that economic development reduces local CO₂ emissions but increases neighbors’ CO₂ emissions. The direct and indirect effects of lnTI are significantly negative and positive, respectively, indicating that technological innovation inhibits CO₂ emissions in the local area but promotes emissions in neighboring areas, respectively. The direct and indirect effects of lnPD are insignificantly and significantly negative, implying that the larger population size increases neighbors’ carbon emissions. As mentioned by Zheng et al. (2014), economic development, population growth, and technological innovation are associated with agglomeration forces that increase local resource efficiency but exacerbate neighbors’ carbon emissions by reducing energy efficiency (Chica-Olmo et al., 2020).

This paper conducts a robustness test by replacing the explained variable and matrix, and the estimation results are presented in Table 5. Column (1) uses total CO₂ emissions rather than per capita CO₂ emissions are the explained variable. The spatial matrix in Column (2) is replaced by the economic distance weight matrix. In Column (3), both the explained variable and the spatial matrix are replaced. As shown in Table 5, the coefficients of lnEGT and WlnEGT are statistically negative and positive, respectively, indicating that economic growth targets are conducive to carbon emission reduction in the local area but not in the neighboring areas. Moreover, the decomposition results indicate that the direct and indirect effects of lnEGT are negative and positive, respectively, which is in line with the baseline regression. For example, the direct and indirect effects of lnEGT are -0.114 and 0.436 in Column (3), respectively, which means that a 1% increase in economic growth targets can result in a 0.114% decrease and 0.436% increase in CO₂ emissions for local and neighboring areas, respectively. In short, the robustness test results are consistent with the baseline regression results, demonstrating the reliability of our findings.

To capture the characteristics of spatial spillover effects, the spatial matrix is reset as follows: The element is recorded as the reciprocal of the geographic distance between two cities if that distance is less than the threshold distance; otherwise, it is recorded as 0. The initial value of threshold distance and progressive distance are both set to 100 km, The threshold distance and progressive distance are both set to 100 km, so the spatial econometric analysis is conducted from 100 km and performed every 100 km thereafter.

Table 6 presents the decomposition results within 1,400 km. The local and spillover effects can be divided into three intervals. The first interval is within 400 km, the direct and indirect effects are statistically negative and positive, respectively, and the inhibiting and promoting effects of economic growth targets on local and neighboring carbon emissions increase with geographic distance. Most cities within 400 km are geographically adjacent or are part of the same province. While they engage in vicious competition for social resources and political promotion, they also participate in cross-regional cooperation to reduce carbon emissions. The second interval ranges between 500 and 1,200 km. The direct and indirect effects of preset targets on CO₂ emissions in local and nearby areas diminish as distance increases. In terms of career advancement, local officials are mainly competing with their peers from their provinces. Consequently, there is little or no strategic interaction between cross-regional cities in setting economic growth targets, which reduces the amplification of preset targets and the competitiveness for liquidity resources. More than 1,200 km is the third interval. The direct effects are irregular, and the indirect effects are no longer significant. This is mainly because most of the elements in the geographical distance weight matrix are 0, which leads to the white noise sequence. In summary, the spatial spillover effect of economic growth targets on CO₂ emissions is attenuated by geographical distance.

To investigate the difference of economic growth target on carbon emissions among cities with different resource endowments, we use per capita education expenditure, deposits and loans to GDP, and fixed-asset investment to GDP as substitutes for human, financial, and material resources, respectively, and then divide the samples into high- and low-resource groups according to the median value for analysis.

We pay more attention to the role of the core variable, namely, economic growth targets. In Columns (1)–(2), the direct effects are significantly negative, indicating that economic growth targets increase CO₂ emissions in cities with different human resources. The indirect effects in Columns (1)–(2) are insignificant and significant, respectively, which means that preset targets in cities with high human resources have no spillover effects, whereas those in cities with low human resources worsen neighbors’ carbon emissions. Carbon emission reduction requires not only well-educated labor forces to develop clean technologies to reduce energy consumption, but also highly skilled workers to put these technologies into practice to improve energy efficiency. Economic growth targets and environmental protection goals put great pressure on many cities, especially those with limited human resources, to compete for highly skilled workers, resulting in a reduction in local carbon emissions and an increase in neighbors’ carbon emissions.

In Columns (3)–(4), the direct effects of lnEGT are statistically negative, suggesting that economic growth targets reduce CO₂ emissions in cities with various financial resources. The indirect effects of lnEGT are significantly positive, which means that local economic growth targets worsen neighbors’ CO₂ emissions. Notably, the direct and indirect effects are higher in cities with low financial resources than in cities with high financial resources. This means that local inhibition and adjacent promotion of preset targets are weaker in cities with high financial resources than in cities with low financial resources. Several initiatives that contribute to carbon emission reduction, including industrial upgrading and technological innovation, are inseparable from financial support. However, under promotion tournaments, local officials may demonstrate their abilities to superiors for career advancement by pursuing high economic growth rates. In turn, this will prompt them to intervene in bank management decisions, such as loan issuance, resulting in less support for carbon emission reduction investments and financing activities.

Third, the direct effects of lnEGT in Columns (5)–(6) are significant and insignificantly negative, respectively, indicating that the economic growth target reduces CO₂ emissions in cities with high material resources, but does not reduce CO₂ emissions in cities with low material resources. The indirect effects of lnEGT are significantly positive in Columns (5)–(6), which means that economic growth targets in the two types of cities increase their neighbors’ CO₂ emissions. These results imply that carbon emissions can only be materially reduced when the infrastructure construction is improved in the fields of environmental protection and carbon markets; that is, material resources are essential for low-carbon development. Overall, the findings not only verify baseline results on a whole, but also indicate that there are heterogeneous effects in cities with differently endowed resources Table 7.

In this section, we reveal we reveal how economic growth targets affect CO₂ emissions. The results with fiscal decentralization as a mediation variable are presented in Columns (1)–(2) of Table 8. The coefficient of lnEGT is significantly positive in Column (1), indicating that economic growth targets contribute to fiscal decentralization. The coefficients of lnFD and WlnFD in Column (2) are negative, although the latter is not statistically significant. Additionally, the coefficients of lnEGT and WlnEGT remain significantly negative and positive after introducing the fiscal decentralization variable. The sign of the product of lnEGT in Column (1) and lnFD in Column (2) is the same as that of lnEGT in Column (2), whereas the sign of the product of lnEGT in Column (1) and WlnFD in Column (2) is opposite to that of WlnEGT in Column (2). These results mean that fiscal decentralization plays a mediating and masking role in linking economic growth targets and CO₂ emissions in local and neighboring regions, respectively. In other words, economic growth targets improve local inhibition and weaken adjacent promotion by promoting fiscal decentralization. For example, the coefficient of lnEGT is 0.153, and the direct and indirect effects of lnFD are -0.117 and -0.314, which means that a 1% increase in fiscal decentralization could reduce CO₂ emissions in local and neighboring regions by 0.018 and 0.048%, respectively.

Columns (3)–(4) show the results of environmental decentralization as a mediation variable. The coefficient of lnEGT is negative and passes the 5% significance test in Column (3), indicating that economic growth targets could hinder environmental decentralization. The coefficients of lnED and WlnED in Column (4) are positive and pass the 1% significance test. Furthermore, after introducing the environmental decentralization variable, the coefficients of lnEGT and WlnEGT remain significantly negative and positive, respectively. Thus, the sign of the product of lnEGT in Column (3) and lnED in Column (4) is the same as that of lnEGT in Column (4), while the sign of the product of lnEGT in Column (3) and WlnED in Column (4) is different from that of WlnEGT in Column (4). These findings imply that environmental decentralization plays a mediating and masking role in the relationship between economic growth targets and local and neighbors’ CO₂ emissions, respectively. That is, environmental decentralization increases the inhibition of economic growth targets on local CO₂ emissions and decreases the promotion of economic growth targets on neighbors’ CO₂ emissions. For instance, the lnEGT coefficient is -0.042, and the direct and indirect effects of lnED are 0.143 and 1.763, indicating that a 1% increase in environmental decentralization could reduce local and neighbors’ CO₂ emissions by 0.006 and 0.074%, respectively.

Based on these findings, we conclude that economic growth targets exert their influence on carbon emissions through fiscal decentralization and environmental decentralization. Specifically, economic growth targets reduce local CO₂ emissions by promoting fiscal decentralization and inhibiting environmental decentralization. Meanwhile, economic growth targets increase neighbors’ CO₂ emissions by improving the negative spillover effect of fiscal decentralization and the positive spillover effect of environmental decentralization. Environmental pollution will not only impact regional economic development by limiting foreign investment and talent introduction, but also restrict local officials’ career development as the central government implements a one-vote veto system for environmental protection. Thus, with sufficient political incentives, local governments will strengthen pollution control while developing the local economy. Fiscal decentralization and environmental decentralization mean that local governments have the initiative to promote coordinated development between the economy and the environment, and they can optimize fiscal expenditure structure and formulate environmental protection policies based on local circumstances, which may reduce pollution more effectively.