The impact of aging and urbanization on CO₂ emission in Chinese cities: an empirical analysis

Abstract

Rapid aging and urbanization pose major challenges to global CO₂ reduction efforts, particularly in China. As such, effective carbon reduction strategies must account for their combined impact on emissions. However, existing research pays insufficient attention to the combined impact of aging and urbanization on CO₂ emissions and the underlying economic drivers. Furthermore, city-level analyses for projecting emissions related to these demographic shifts remain scarce. To address these limitations, this study employs an STIRPAT model using city-level data from China for 2010–2020 to estimate the combined effects of population aging and urbanization on CO₂ emissions. Then, we explore the economic drivers underlying this relationship. Finally, we project city-level CO₂ emissions driven by population aging and urbanization under different shared socioeconomic pathways (SSPs) in 2030–2060. Our findings suggest that: (a) The emission-increasing effect of aging outweighs the mitigating impact of urbanization, and together these opposing forces contribute to the overall rise in emissions. (b) Per capita disposable income and the size of the tertiary sector are key economic drivers of population aging, contributing to increased CO₂ emissions. The tertiary sector also significantly influences urbanization, which in turn facilitates emission reductions. (c) The relative contributions of CO₂ emissions from population aging and urbanization are similar under SSP1, SSP4, and SSP5. However, the absolute levels of CO₂ emissions resulting from the combined effects of aging and urbanization exhibit significant variation across cities under the SSP1–SSP5 scenarios. Based on the results, we offer policy recommendations to support carbon mitigation efforts.

Introduction

Rising global aging and urbanization pose significant challenges to carbon reductions and achieve Paris Agreement targets. According to the World Population Prospects 2022¹, the global population aged 65 and above is increasing faster than those under 65, and is projected to rise from 10% in 2022 to 16% by 2050. Meanwhile, the World Urbanization Prospects 2018² reports that 55% of the global population are urban dwellers, and projects to reach 68% by 2050, with most growth concentrated in Asia and Africa. China is one of the fastest-aging and urbanizing countries in the world over the past and coming decade^3,4, with these shifts posing significant challenges to achieve carbon neutrality by 2060. It is crucial for China to clarify the combined effects of population aging and urbanization on CO₂ emissions and to formulate targeted mitigation strategies in response.

The UN’s New Urban Agenda focuses on the interlinked challenges of aging urban populations⁵, which manifest through demographic structure, economic behavior, and public resource allocation. This perspective frames aging and urbanization as presenting both a challenge and an opportunity for implementing the SDGs. Aging poses challenges for meeting SDG3 (Good Health and Wellbeing) by straining health and long-term care systems, while simultaneously driving urban service sector growth through consumption shifts and increased healthcare demand⁶, a process accelerated by rapid urbanization. Rural-to-urban youth migration diverges regional aging patterns, creating “rural hollowing”⁷ and decelerating urban aging⁸. Concurrently, the process of urbanization in turn shapes the trajectory of aging. This shaping effect is reflected in the urban social sector⁹, driving increased demand for enhanced healthcare and social security for the aging population. Existing literature shows that projections which account for both aging and urbanization are 1.7–6.7% fewer in terms of excess temperature-related cardiovascular deaths (2020–2039) than those considering aging alone¹⁰. These interactions underscore the need to assess the combined effect of aging and urbanization on CO₂ emissions and to identify related sustainability challenges and opportunities.

However, existing studies have predominantly examined the direct impacts of population aging or urbanization on CO₂ emissions in isolation^6,11,12, while few have explored their combined environmental effects by integrated approaches¹³. Furthermore, despite extensive research on aging’s socioeconomic impacts at household and national scales¹⁴, little attention has been given to the prefectural-city level¹⁵, which is a principal locus for the future demographic transitions shaping our future¹⁶.

While the impacts of population aging and urbanization on CO₂ emission are largely driven by regional economic conditions¹⁷, limited research has explored the specific economic factors that influence these demographic trends and, in turn, affect CO₂ emissions. Existing studies on economic-driven factors have shown that lower-income elderly populations tend to adopt less environmentally friendly consumption habits, resulting in higher CO₂ emissions⁶. Conversely, wealthier households, with greater financial resources and choices, prefer to use cleaner energy sources like electricity and natural gas¹⁸. Additionally, Chikaraishi et al.¹⁹ found that higher per capita GDP and a greater service-sector share contribute to more environmentally sustainable urbanization. Collectively, these findings highlight the academic importance of exploring the economic drivers of demographic changes and the resulting impacts of those changes on CO₂ emissions, as closing this research gap is essential for informing effective, economically grounded carbon reduction policies. Furthermore, while existing studies have predicted future CO₂ emissions trends related to population aging and urbanization at the national²⁰ or provincial level²¹, they often overlook city-level forecasts, which limits the ability to reveal regional differences in China’s future emissions patterns.

This study empirically examines the combined effects of population aging and urbanization on CO₂ emissions by (a) constructing a city-level STIRPAT model that accounts for the endogeneity of aging and urbanization; (b) applying the Control Function Approach (CFA) to analyze the influence of economic factors; and (c) projecting city-level CO₂ emissions in China from 2030 to 2060 under the five SSP scenarios (SSP1–SSP5) defined in the IPCC Sixth Assessment Report²².

Results

CO₂ emissions impacts of aging and urbanization

Using city-level data from 2010 to 2020 in China, we conduct an empirical analysis of the combined effects of population aging and urbanization on CO₂ emissions. The results (Table 1) show that, China’s CO₂ emissions exhibited an upward trend, with aging and urbanization exert opposing effects. Specifically, the results obtained from the instrumental variable (IV) approach indicate a positive relationship between population aging (PAG) and CO₂ emissions, and a negative relationship between urbanization (URB) and CO₂ emissions. Table 1 reports the estimates under both exogeneity and endogeneity assumptions, using fixed effects, 2SLS, 2SPS, and 2SRI models. Based on the test outcomes, we select the 2SRI model for detailed regression analysis. Supplementary Table 4 presents the detailed results of tests for multicollinearity, endogeneity, and the validity of the instrumental variables.

Table 1 Comparative regression results of aging and urbanization on CO₂ emissions

The 2SRI regression results in column (4) of Table 1 show that PAG is significantly positively associated with CO₂ emissions at the 1% significance level, with a one-unit increase in aging expected to raise CO₂ emissions by 3.007 units (e^1.101 ≈ 3.016). URB is significantly negatively associated with CO₂ emissions at the 5% significance level, with a one-unit increase in urbanization expected to reduce emissions by 0.409 units (e^−0.893 ≈ 0.409). Column (4) shows that the first-stage residuals for PAG and URB are statistically significant in explaining CO₂ emissions at the 1% and 10% levels, indicating the validity of the instrumental variables and the appropriateness of the 2SRI estimation method. The finding that population aging is positively associated with CO₂ emissions is consistent with Yu et al.¹². Similarly, the negative association between urbanization and CO₂ emissions is consistent with Ala-Mantila et al.²³, which argues that, given equal levels of consumption and expenditure, urban lifestyles result in less environmental damage than rural ones.

The statistically significant positive coefficient of the aging-urbanization interaction term in column (5) demonstrates is synergistic effect on CO₂ emissions. This empirical evidence indicates that population aging progressively offsets the CO₂ reduction benefit of urbanization, potentially even reversing the relationship between urbanization and CO₂ emissions from negative to positive, thereby hindering progress toward emission reduction goals.

The influence of economic factors on aging, urbanization, and CO₂ emissions

The existing literature suggested that regional economic development is a key driver of population aging and urbanization²⁴. We further employ the CFA method to examine how economic factors drive the effects of aging and urbanization on CO₂ emissions in China.

As illustrated in Table 2a, columns (1)–(4) present the first-stage regression results, showing that fiscal revenue (REV), per capita disposable income (PINC), and the share of the tertiary sector (TER) are all significantly associated with PAG. Specifically, REV is significantly negatively correlated with PAG at the 1% significance level, while PINC is significantly negatively correlated with PAG at the 5% significance level. A possible explanation is that higher fiscal revenue and per capita disposable income enhance a city’s attractiveness to migrant workers, leading to an inflow of labor that reduces the proportion of the elderly population. In contrast, TER is significantly positively correlated with PAG at the 1% significance level. According to the Petty-Clark theorem²⁵, as economic development progresses and per capita national income increases, the service sector gradually becomes the primary driver of economic growth. The empirical results suggest that during economic growth and industrial restructuring, if the total urban population remains constant, population aging will increase alongside the expansion of the service sector. Column (6) shows that PAG driven by PINC is significantly positively correlated with CO₂ emissions at the 1% significance level. Column (7) indicates that PAG driven by TER is also significantly positively correlated with CO₂ emissions at the 1% significance level.

Table 2 Estimation results for economic factors using the CFA method

As illustrated in Table 2b, columns (1) and (2) indicate that neither REV nor PINC is significantly correlated with URB. Columns (3)–(4) show that TER and fixed asset investment (INV) are both significantly negatively correlated with URB at the 1% significance level. Column (5) indicates that URB associated with TER is significantly negatively correlated with CO₂ emissions at the 5% significance level. Based on these findings, this study concludes that per capita disposable income and the size of the tertiary sector are key drivers of population aging, which in turn is significantly positively correlated with CO₂ emissions. Additionally, the size of the tertiary sector is an important driver of urbanization, which in turn is significantly negatively associated with CO₂ emissions.

Aging and urbanization-related CO₂ emissions from 2030 to 2060

The projected rapid growth of population aging and urbanization under all scenarios²⁶ is expected to pose significant challenges for China in achieving its carbon neutrality goal. Based on the coefficient estimation results from the empirical model, this study projects China’s future CO₂ emissions for 2030–2060 at the city level under different SSP scenarios. With a mean absolute percentage error (MAPE) of 19%, which falls within the 10–20% range defined by Pao and Tsai²⁷ as indicative of good forecast accuracy, the use of STIRPAT regression coefficients to predict CO₂ emissions for 2030–2060 is justified.

The combined effects of population aging and urbanization contribute to the overall growth of CO₂ emissions, with the positive impact of population aging on emissions in all projected scenarios outweighing the negative impact of urbanization. Among 200 Chinese cities, SSP3 records the lowest CO₂ emissions influenced by the combined effects of population aging and urbanization, at 4663.29 Mt in 2030. In contrast, SSP5 shows the highest emissions influenced by the combined effects, reaching 5807.18 Mt in 2030. Supplementary Table S5 reports the projected total CO₂ emissions in detail. Among all scenarios, Fig. 1 shows that SSP3 also has the lowest share of emissions influenced by the combined effects relative to total projected emissions, at 30.86% in 2030 and 48.32% in 2060, followed by SSP2, SSP4, and SSP5, while SSP1 has the highest proportion. Under SSP1, SSP4, and SSP5, the proportion of CO₂ emissions attributable to aging and urbanization is relatively similar, reaching 30.02%, 30.22%, and 29.94% respectively in 2030, and increasing significantly to 46.19%, 47.36%, and 45.43% by 2060. SSP1 exhibits the highest degree of population aging among all SSP scenarios²⁶, with the positive impact of population aging on CO₂ emissions reaching 39.72% in 2030 and rising to 70.29% by 2060, indicating the greatest pressure for carbon emission growth. Conversely, the negative impact of urbanization is most pronounced under SSP5, accounting for 9.73% in 2030 and 24.46% in 2060, slightly higher than that under SSP1 and SSP4. The highest level of urbanization among all scenarios²⁶ reflects the greatest potential for emission reduction.

Fig. 1: Proportions of CO₂ emissions associated with population aging and urbanization in total CO₂ emissions.

The impact of population aging is represented by the proportion of CO₂ emissions related to it in the total CO₂ emissions, shown as blue bars. Given the positive relationship between population aging and CO₂ emissions, the related emissions are greater than zero. The impact of urbanization is represented by the proportion of CO₂ emissions related to it in the total CO₂ emissions, shown as red bars. Given the negative relationship between urbanization and CO₂ emissions, the related emissions are less than zero.

City-level CO₂ emissions disparities in 2060

Although the relative contribution of CO₂ emissions associated with population aging and urbanization in total CO₂ emissions shows similar values across SSP1, SSP4, and SSP5, this study further examines the absolute differences under various scenarios. Our findings suggest that the combined effects of aging and urbanization will lead to substantial variation in future CO₂ emissions across cities, with marked regional disparities. Overall (Fig. 2), in 2060, CO₂ emissions associated with these demographic shifts are projected to be lower in cities along the southeastern coast and higher in cities located in the northwest and northeast regions. Under the SSP2 scenario, 35 cities, including Taiyuan, Zibo, and Weifang, fall into the lowest range of CO₂ emissions, where future CO₂ emissions are expected to be more influenced by urbanization than by population aging, suggesting a greater potential for achieving carbon reduction targets compared to other cities. Conversely, CO₂ emissions in the highest range under the SSP2 scenario are recorded in cities such as Changzhou, Yan’an, and Shenyang, among a total of 40 cities. In these cities, future CO₂ emissions are expected to be more influenced by population aging than by urbanization, leading to greater carbon emission pressures compared to other cities.

Fig. 2: CO₂ emissions associated with population aging and urbanization in 2060.

Panels (a) to (e) in Fig. 2 present the projected CO₂ emissions associated with population aging and urbanization in 2060 for SSP1 to SSP5, respectively. The results are proportionately categorized into five levels. Specifically, the dark green level represents the smallest combined effects of population aging and urbanization on total CO₂ emissions, followed by light green, yellow, and orange. The bright red level indicates the highest combined effects. Areas in white on the map indicate regions with no available data.

Compared to the SSP2 scenario, some cities under the SSP1 scenario, such as Hengyang, Yingtan, and Linyi, are expected to transition from lower to higher emission levels, among a total of 39 cities. These cities are primarily located in central and northern China, suggesting that the green development pathway of SSP1 has improved living standards and life expectancy, which in turn has accelerated population aging and increased CO₂ emissions. Moreover, the projected CO₂ emissions under SSP1 in Lianyungang, Yancheng, Zhenjiang, and Yangzhou have decreased compared to those under SSP2, with these cities shifting from higher emission levels under SSP2 to lower ones under SSP1. This indicates that SSP1 has enhanced urban attractiveness, facilitated rural-to-urban migration, accelerated urbanization, and consequently reduced projected CO₂ emissions.

Under the SSP3 scenario, the overall level of urbanization is higher compared to SSP2, leading to lower projected CO₂ emissions. Notably, no cities experience a transition from lower to higher emission levels under SSP3 relative to SSP2. A total of 137 cities, including Jian, Suqian, and Shaoyang, are projected to shift from higher CO₂ emissions levels under SSP2 to lower levels under SSP3. These cities are primarily located in eastern and central China, indicating that despite intensified regional competition and resource constraints under SSP3, they remain relatively attractive to populations, which in turn drives deeper urbanization and contributes to lower CO₂ emissions.

In most cities, projected CO₂ emissions under SSP4 are higher than under SSP2. Fifty cities, including Suzhou, Xiangxi, and Suizhou, are expected to shift from lower emission levels under SSP2 to higher levels under SSP4. In these cities, population aging is greater compared to that under SSP2. Due to the regional disparities under SSP4, projected CO₂ emissions in many cities are lower than under SSP2. Fourteen cities, such as Beihai, Bozhou, and Laibin, are expected to shift from higher emission categories under SSP2 to lower ones under SSP4. This indicates that these cities may experience higher levels of urbanization than those under SSP2, contributing to future reductions in CO₂ emissions associated with both population aging and urbanization.

The total projected CO₂ emissions under SSP5 are higher than those under SSP2. This indicates that in the fossil fuel-driven, high-growth pathway of SSP5, the emission increases associated with population aging outweigh the reductions achieved through urbanization. Seventy-five cities, including Huaihua, Xiangxi, and Qiandongnan, are projected to have higher CO₂ emissions under SSP5 than under SSP2, with a corresponding shift to higher emission categories. Meanwhile, a few cities show lower emissions under SSP5 compared to SSP2. These cities have experienced a significant influx of rural migrants, resulting in rapid urban population growth and a lower share of elderly residents. However, the resulting changes in CO₂ emissions are not substantial enough to shift their emission categories.

Discussion

This paper integrates population aging and urbanization into a unified model, setting it apart from previous studies that analyze them separately or rely only on elderly samples to assess the impact of urbanization. We account for potential endogeneity between these two demographic shift factors and CO₂ emissions, to obtain consistent estimates of the combined effects of aging and urbanization. The estimation results indicate that the positive effect of population aging on CO₂ emissions exceeds the negative effect of urbanization, suggesting that the accelerating aging process is reshaping the emissions structure and diminishing the carbon reduction potential traditionally attributed to urbanization.

There is a significant positive correlation between population aging and CO₂ emissions in Chinese cities, and this relationship is primarily driven by per capita disposable income and the share of the tertiary sector. As physical function declines with age, the elderly require higher indoor temperatures to ensure a healthy and comfortable living environment²⁸. In China, coal remains the dominant fuel for household heating, and its use rises with the growing elderly population, contributing to higher carbon emissions associated with population aging. Moreover, nostalgia and conservative consumption habits reduce the elderly’s motivation to purchase energy-efficient products²⁹. If the elderly population lives in older buildings with outdated infrastructure, it can further lead to inefficient use of energy for heating and cooling, thereby contributing to higher CO₂ emissions. Rising urban incomes attract more migrant labor, leading to population growth and a decline in the aging rate. However, rising income levels do not directly alter the consumption patterns of the elderly. Instead, the combined effect of higher incomes and greater purchasing power further intensifies the CO₂ emissions associated with population aging. On the other hand, the growth of tertiary industries such as technology and information services can support medical innovation and improve elderly care. This not only extends life expectancy and accelerates the aging trend but also stimulates new consumption patterns among the elderly, further contributing to rising CO₂ emissions pressures. As aging and urbanization increasingly interact in China, the effectiveness of traditional emission reduction strategies that focus solely on urbanization diminishes. Consequently, policymakers must integrate demographic factors into policy design and explore alternative approaches to emission reduction. This study suggests that governments promote low-carbon awareness among the elderly and encourage the adoption of energy-efficient products and services. Specific measures include increasing subsidies for low-carbon household appliances and expanding the range of eligible subsidized products, while also strengthening efforts to promote these low-carbon consumption activities among the elderly population. Meanwhile, governments and enterprises should enhance urban appeal to highly skilled labor by reasonably adjusting wage structures and steadily increasing income levels.

China’s urbanization exhibited a significant negative correlation with CO₂ emissions, and this relationship is primarily driven by the share of the tertiary sector. We argue that urbanization affects emissions through both production and consumption behavior. On one hand, urbanization is accompanied by advancements in human capital and technology, which enhance energy efficiency and help to reduce CO₂ emissions³⁰. On the other hand, given urban residents’ strong willingness to adopt low-carbon products and their widespread use of green appliances powered by renewable energy, further urbanization can effectively promote green consumption, thereby helping to reduce CO₂ emissions³¹. The growth of the tertiary sector generates new urban employment opportunities³² and typically involves lower energy consumption intensity. However, the shift toward industries distinct from labor-intensive or traditional agricultural sectors poses challenges for rural migrants entering cities, particularly regarding the skills and learning capacity required. This mismatch may lead to labor supply instability in the tertiary sector and, consequently, limit urban attractiveness to the rural population in China. Therefore, this study does not reveal a statistically significant positive correlation between the proportion of the tertiary sector and urbanization. Based on these findings, governments should offer skills training to support rural laborers in meeting the demands of emerging job opportunities, thereby improving the alignment between labor supply and market demand. At the same time, policy guidance on energy conservation for certain energy-intensive industries within the tertiary sector should be strengthened, with increased investment in energy-saving technologies and strict penalties imposed for illegal energy use.

From 2030 to 2060, the proportion of CO₂ emissions attributable to population aging in Chinese cities is projected to exceed the proportion linked to urbanization. The positive effect of population aging on emissions and the negative effect of urbanization are both expected to intensify over time. The relative contribution of CO₂ emissions associated with population aging and urbanization in total CO₂ emissions remains similar across SSP1, SSP4, and SSP5. However, the absolute impact of the combined effects of aging and urbanization is projected to cause substantial variation in future CO₂ emissions across cities. By 2060, cities with lower predicted CO₂ emissions related to population aging and urbanization will be primarily concentrated in central and southeastern China, while those with medium to high emissions will be mainly located in the northern and northeastern regions. The future trajectory of CO₂ emissions associated with population aging and urbanization will be shaped by the characteristics of projected scenarios, as well as by existing demographic and urban development patterns across cities. SSP1 and SSP5 exhibit significant aging trends, resulting in higher projected CO₂ emissions compared to SSP2. In contrast, SSP3 features accelerated urbanization, leading to lower emissions than SSP2. SSP4 presents a more complex picture, as some cities are projected to exceed SSP2 in emissions while others may fall below, reflecting a highly uneven distribution.

The study has certain limitations that should be noted. CO₂ emissions projections for 2030–2060 related to population aging and urbanization are based on regression coefficients derived from data spanning 2010 to 2020. This method does not account for cohort differences in savings and consumption preferences⁶. In particular, earlier-born elderly populations generally display lower consumption and higher savings tendencies, which is reflected in lower historical regression coefficients. Consequently, applying these historical coefficients may lead to an underestimation of future CO₂ emissions associated with aging. Future research should further identify and analyze the evolving behavioral patterns of different elderly cohorts to improve the accuracy of long-term emissions projections.

Methods

Basic model foundation

This study employs the STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) model, originally developed by Dietz and Rosa³³ based on the IPAT equation³⁴ (Eq. 1), to empirically examine the correlation between aging, urbanization, and CO₂ emissions. Compared to other models, the STIRPAT model is more suitable for examining the intrinsic relationship between human activities and the environment, as it offers greater flexibility to expand the model by adding, removing, or decomposing variables, thereby enhancing its analytical and explanatory capacity³⁵. To ensure data stationarity and facilitate computation, we transform the basic STIRPAT model into a logarithmic form³⁶ (Eq. 2). Urbanization is then incorporated into the basic STIRPAT model (Eq. 3) to assess the combined impact with aging on CO₂ emissions.

$$I=a{P}^{{b}}{A}^{{c}}{T}^{{d}}e$$

(1)

$${\text{ln}}\,I={\text{ln}}\,a+b\,{\text{ln}}\,{\rm{P}}+c\,{\text{ln}}\,A+d\,{\text{ln}}\,T+{\text{ln}}\,e$$

(2)

In line with the research focus, CO₂ emissions are selected as the indicator of environmental impact (I). Aging and urbanization are selected as the indicators of population (P) based on both their significant impact on CO₂ emissions and their representation of key demographic dimensions, namely age structure and spatial distribution. Per capita GDP is selected as the indicator of affluence (A) for three reasons: its status as a standard measure of regional economic welfare; its widespread use in prior literature^37,38, which ensures comparability; and its function as a per capita measure that controls for population size, thereby mitigating multicollinearity with population (P) and isolating the effect of affluence. Technological progress (T) is conceptualized across three dimensions²⁰: average years of education per capita captures the knowledge-innovation capacity³⁹, reflecting the “soft power” for decarbonization rooted in aging and urbanization; energy intensity indicates efficiency improvements in economic production from technological upgrades; and energy structure reveals the technical shift and societal preference in end-use consumption, signifying both fossil fuel dependency and public drive for cleaner energy. Moreover, the literature commonly employs energy intensity and energy structure to denote technological progress^31,40. In summary, our variable selection in the STIRPAT model draws on existing research but is tailored to the core explanatory variables of aging and urbanization, thereby enhancing estimation accuracy.

$${\text{ln}}\,C{O}_{{2}}={\text{ln}}\,a+{b}_{1}\,{\text{ln}}\,PAG+{b}_{2}\,{\text{ln}}\,URB+{b}_{3}\,{\text{ln}}\,PGDP+{b}_{4}\,{\text{ln}}\,EDU+{b}_{5}\,{\text{ln}}\,STR+{b}_{6}\,{\text{ln}}\,INS+{\text{ln}}\,e$$

(3)

Where I, P, A, and T represent the environmental impact, population, affluence, and technological progress, respectively. Here, CO₂, PAG, URB, PGDP, EDU, STR, and INS represent CO₂ emissions, population aging, urbanization, per capita GDP, average years of education per capita, energy structure, and energy intensity, respectively. a, b, c, and d are parameters to be estimated, and e represents the random disturbance term.

Assessment of the coefficient

To obtain consistent, unbiased, and efficient estimates, this study employs the Instrumental Variables (IV) method to address endogeneity issues. A valid instrumental variable must satisfy three conditions: (a) it must be strongly correlated with the endogenous variables, (b) it must be exogenous to the model, meaning uncorrelated with the error term, and (c) the number of instrumental variables must be at least equal to the number of endogenous variables. This study selects the gender ratio (GEN) as the instrumental variable for PAG. GEN, defined as the number of males per 100 females, increases with a rise in the male population. It influences population size by affecting marriage and fertility rates, as well as migration driven by employment opportunities or marital considerations⁴¹. Moreover, GEN has no significant direct impact on CO₂ emissions, reinforcing its suitability as an instrument. This study uses the share of the tertiary sector in GDP (TER) as the instrumental variable for URB. Growth in the tertiary industry stimulates urban economic spillover effects, enhancing the attractiveness of the region to external factors of production and thereby promoting urbanization⁴². However, compared to the secondary sector, the tertiary sector contributes less directly to CO₂ emissions. Instead, it influences emissions indirectly by promoting industrial upgrading and technological progress, making it a suitable instrumental variable for urbanization⁴³.

This study employs the two-stage least squares (2SLS) and two-stage predictor substitution (2SPS) methods to test the validity of the instrumental variables⁴⁴. Then, we apply the two-stage residual inclusion (2SRI) methods to estimate the relationship between PAG, URB, and CO₂ emissions. The 2SLS method addresses endogeneity in linear models through a two-stage regression process. In the first stage, the endogenous variables are regressed on the instrumental variables using a linear regression model. We apply a fixed-effects model with city-clustered robust standard errors for this regression (Eq. 4). In the second stage, the dependent variable is regressed on the fitted values of the endogenous variables from the first stage, along with the remaining exogenous variables (Eq. 5).

$${\text{ln}}\,{X}_{e}={\text{ln}}\,{a}_{1}+{b}_{1}{\text{ln}}\,Z+{b}_{2}\,{\text{ln}}\,{X}_{{\rm{o}}}+{\text{ln}}\,{e}_{1}$$

(4)

$${\text{ln}}\,C{O}_{2}={\text{ln}}\,{a}_{2}+{b}_{3}\,{\text{ln}}\,{\hat{{X}}}_{{e}}+{b}_{4}\,{\text{ln}}\,{X}_{{\rm{o}}}+{\text{ln}}\,{e}_{2}$$

(5)

Where, X_e represents the endogenous variables, defined as X_e = [PAG URB]. Z represents the instrumental variables, defined as Z = [GEN TER]. X_o represents the exogenous control variables, defined as X_o = [PGDP EDU STR INS].

The 2SPS method is an extension of the 2SLS method for nonlinear models⁴⁵. In the first stage, the endogenous variables are regressed nonlinearly on the instrumental variables (Eq. 6). In the second stage, the dependent variable is regressed nonlinearly on the fitted values of the endogenous variables from the first stage, along with the remaining exogenous variables (Eq. 7).

$${\text{ln}}\,{X}_{e}=M\left({\text{ln}}\,{a}_{1}+{b}_{1}\,{\text{ln}}\,Z+{b}_{2}\,{\text{ln}}\,{X}_{o}+{\text{ln}}\,{e}_{1}\right)$$

(6)

$${\text{ln}}\,C{O}_{2}=M\left({\text{ln}}\,{a}_{2}+{b}_{3}\,{\text{ln}}\,{\hat{{X}}}_{{e}}+{b}_{4}\,{\text{ln}}\,{X}_{{o}}+{\text{ln}}\,{e}_{2}\right)$$

(7)

We utilize Nonlinear Least Squares (NLS) for the regression analysis in this section. Here, M(.) represents the nonlinear model, while the definitions of the other symbols remain consistent with those provided earlier.

The 2SRI method was first introduced by Hausman⁴⁶. Terza et al.⁴⁷ demonstrated through mathematical modeling and data simulation that the 2SRI helps correct estimation bias caused by endogenous variables. In the first stage, the endogenous variables are regressed nonlinearly on the instrumental variables (Eq. 6). In the second stage, the dependent variable is regressed nonlinearly on the fitted residuals from the first stage, along with the remaining exogenous variables (Eqs. 8, 9).

$${\text{ln}}\,{\hat{e}}_{1}={\text{ln}}\,{X}_{{e}}-{\text{ln}}\,{a}_{1}-{b}_{1}\,{\text{ln}}\,Z-{b}_{2}\,{\text{ln}}\,{X}_{{o}}$$

(8)

$${\text{ln}}\,C{O}_{2}={\text{ln}}\,{a}_{2}+{b}_{3}\,{\text{ln}}\,{X}_{{e}}+{b}_{4}\,{\text{ln}}\,{X}_{{o}}+{b}_{5}\,{\text{ln}}\,{\hat{e}}_{1}+{\text{ln}}\,{e}_{2}$$

(9)

This section also employs NLS for coefficient estimation. Where $\mathrm{ln}{\hat{e}}_{1}$ represents the fitted residual values. The coefficient b₅ is used to test whether there is a significant difference in the estimated coefficient (b₃) of the endogenous variable between models that control for $\mathrm{ln}{\hat{e}}_{1}$ and those that do not⁴⁸. The specific testing process is outlined in Eqs. 10 and 11. The relationship between the endogenous variable, the residual term, the instrumental variables, and their covariates is formalized in Eq. 10.

$${\text{ln}}\,{X}_{{e}}={c}_{1}\,{\text{ln}}\,Z+{c}_{2}\,{\text{ln}}\,W+{\text{ln}}\,\omega$$

(10)

The residual term in Eq. 9 can be expressed by Eq. 11.

$${\text{ln}}\,e={b}_{5}\left({c}_{2}\,{\text{ln}}\,W+{\text{ln}}\,\omega \right)+{\text{ln}}\,{e}_{2}$$

(11)

Where, W represents the covariates of the instrumental variables, satisfying $E\left({Z}^{{\prime} }W\right)\ne 0$. ω denotes the residual term. Besides the instrumental variable Z, other related variables jointly influence the endogenous variable. As a result, the estimated coefficient c₁ in Eq. 10 differs from the estimated coefficient b₁ in Eq. 4, leading to estimation bias and causing b₃ and b₅ in Eq. 9 to approach zero. Conversely, if b₃ and b₅ do not approach zero, meaning that the estimation results are both statistically significant, they are considered to downward-bound⁴⁹. In other words, when both b₃ and b₅ are statistically significant, the instrumental variables can be considered to meet the selection criteria.

Control Function Approach (CFA)

This study analyzes the economic factors that are significantly associated with PAG and URB by using the control function approach (CFA) from income and expenditure dimensions. Income-related factors include fiscal revenue (REV) and per capita disposable income (PINC), both of which are associated with regional labor force agglomeration, as well as the growth of corporate and labor earnings⁵⁰. These factors influence aging and urbanization processes by attracting population inflows, ultimately impacting CO₂ emissions. Expenditure-related factors include the share of the tertiary sector in GDP (TER) and fixed-asset investment (INV), both of which are linked to local government investment in public service infrastructure, improvements in public infrastructure, enhanced public services, and greater business attractiveness. These processes alter aging and urbanization by improving the quality of life for the elderly and promoting industrial upgrading, ultimately impacting CO₂ emissions.

The CFA method in this study is used to examine the relationship between the explanatory variables within the STIRPAT model and external exogenous variables⁴⁸. In the first stage, PAG and URB are separately regressed on economic factors (Eq. 12). In the second stage, CO₂ emissions are regressed on the fitted residuals from the first stage (Eq. 13). Since other economic factors are highly correlated with PGDP, which may cause multicollinearity, PGDP is excluded from the model in this section.

$${\text{ln}}\,P={\text{ln}}\,{a}_{1}+{b}_{1}\,{\text{ln}}\,ECO+{b}_{2}\,{\text{ln}}\,T+{\text{ln}}\,{e}_{1}$$

(12)

$${\text{ln}}\,C{O}_{2}={\text{ln}}\,{a}_{2}+{b}_{3}\,{\text{ln}}\,P+{b}_{4}\,{\text{ln}}\,T+{b}_{5}\,{\text{ln}}\,{\hat{{e}}}_{1}+{\text{ln}}\,{e}_{2}$$

(13)

Where, P represents population aging and urbanization, defined as P = [PAG URB]. ECO represents economic factors, defined as ECO = [REV PINC TER INV]. T represents technological progress factors, defined as T = [EDU STR INS]. The coefficient b₅ indicates the correlation between population aging or urbanization and CO₂ emissions under the influence of specific economic variables.

Multi-scenario CO₂ emissions projection

Based on the Shared Socioeconomic Pathways (SSPs) proposed in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6)²², this study further projects CO₂ emissions in Chinese cities from 2030 to 2060 under the SSP1-SSP5 scenarios. Specifically, SSP1 represents a sustainable development pathway, characterized by effective control of population growth and improvements in overall population quality. SSP2 reflects a continuation of current policies, with both population growth and economic development following existing trends. SSP3 represents a regional rivalry pathway, characterized by a growing total population and a higher birth rate compared to other scenarios, resulting in a lower proportion of elderly individuals. SSP4 describes an inequality pathway, in which regional disparities lead to uneven patterns of population aging and urbanization. SSP5 represents a fossil fuel-driven development pathway, characterized by relatively stable population growth alongside rapid, balanced urbanization and population aging.

Based on the estimated coefficients in Eq. 3 and projected variable values, this study projects CO₂ emissions in Chinese cities from 2030 to 2060. To assess the combined effects of aging and urbanization on CO₂ emissions, we subtract the CO₂ emissions estimated under the assumption that population aging and urbanization remain at their 2020 levels (with all other variables set to their 2060 projections) from the emissions estimated using 2060 projections for all variables.

Statistics and reproducibility

This study uses a sample of 200 observations for statistical analysis (n = 200). Descriptive statistics of the dataset, including the mean, standard deviation, median, maximum, and minimum values, are presented in Appendix Table S3.

The study period of 2010–2020 aligns with a critical phase in China’s demographic transition. Robustness checks in Appendix Table S7 ensure the reliability of the inferred causal relationships between demographic shifts and CO₂ emissions. The following section explains the theoretical and policy basis for the period selection. (1) The Sixth National Population Census in 2010 revealed an accelerating pace of population aging in China, capturing the entry of the baby boom cohort⁵¹ into old age, which intensified aging and produced a new, mobility-influenced structural pattern. Our findings provide important insights for addressing the environmental challenges of China’s accelerated aging, which intensified in 2021, four years ahead of projections. (2) The 18th National Congress of the Communist Party of China endorsed people-centered new-type urbanization, marking a systemic transition in the development model⁵². China’s urbanization has shifted from rapid growth (1996–2011) to a quality-oriented model since 2012, characterized by a significantly slower pace of development that began around 2010. Our sample period begins in 2010 to capture the initial, critical transition stage of urbanization. (3) As a national strategy, the 12th Five-Year Plan (2011–2015) maintained the energy intensity assessment and incorporated new targets for the share of fossil energy consumption and carbon intensity controls. These carbon intensity objectives were then raised in the 13th Five-Year Plan (2016–2020). We employ energy structure and intensity data from the 12th and 13th Five-Year Plan periods, which align with a crucial evolution in China’s carbon strategy, thus providing a coherent policy context for population-carbon emissions analysis.

This study conducts a series of statistical tests to evaluate the variable settings and coefficient estimates of the empirical model. Specifically, multicollinearity is assessed using the mean Variance Inflation Factor (VIF). The Hausman test is applied to determine the appropriateness of the Fixed Effects Model (FE) versus the Random Effects Model (RE) and to examine the endogeneity of population aging and urbanization. The Kleibergen-Paap rk LM statistic is used to assess underidentification of the instrumental variables, while the Cragg-Donald Wald F statistic and the Kleibergen-Paap rk Wald F statistic are employed to test for weak identification. As only one instrumental variable is specified for both aging and urbanization, the model is exactly identified, and an overidentification test is therefore unnecessary. Detailed test results are reported in Appendix Table S4.

This study sets the significance levels (α) at 0.01, 0.05, and 0.1. Two-tailed tests are conducted using a non-directional alternative hypothesis. The t-test or z-test was used to generate the P value. If the P value < α, the null hypothesis is rejected, indicating that the independent variable has a statistically significant impact on the dependent variable. Conversely, if the P value > α, we fail to reject the null hypothesis, suggesting that the independent variable does not have a statistically significant effect on the dependent variable. Detailed results of the significance tests and the actual P values are provided in Tables 1 and 2.