Effects of population flow on regional carbon emissions: evidence from China

Population flow can affect regional carbon emissions. Based on the analysis of the dual transmission mechanism of population flow and its effect on carbon emissions, this paper empirically studies the impact of population flow and other related factors on China’s carbon emissions through panel econometric regression and heterogeneity analysis with fixed effect model. The results show that, firstly, in the long or short term, China’s population flow can reduce the growth of carbon emissions. Secondly, the regional population aging and knowledge structure improvement caused by population flow are helpful to reduce carbon emissions, while the regional urbanization improvement caused by population flow is not significantly correlated with the growth of household miniaturization on carbon emissions. Thirdly, from the perspective of heterogeneous geographical divisions, population flow promotes the increase of carbon emissions in the northwest region of the Hu Huanyong Line (Hu Line), while it is opposite in the southeast region of Hu Line. Fourthly, China’s consumption level, per capita GDP, energy intensity, and energy consumption structure have contributed to the growth of carbon emissions, while carbon intensity has a negative effect on carbon emissions. Finally, this paper puts forward relevant suggestions from the perspective of coordinating population policy and energy conservation and emission reduction policy.


Introduction
Recently, global warming caused by carbon emissions growth has become the focus of academic community. Countries make great efforts to accelerate the development of low-carbon economy and achieve the goal of carbon emissions reduction. As the world's second largest economy, China has huge pressure to reduce carbon emissions. In response to this, in 2015, China promised in the Paris Agreement to strive to achieve the goal of reducing carbon emission intensity by 60-65% by 2030 compared with 2005. On March 2021, the Chinese government announced that it would achieve a carbon peak by 2030 and strive to achieve carbon neutralization by 2060. Therefore, it is of great practical significance to study the relevant factors affecting carbon emissions providing theoretical and empirical support for the formulation of energy conservation and emission reduction policies.
Currently, there is plenty of research on the factors affecting carbon emissions. Among which, economic development level, population factor, technology level, urbanization, energy structure, and industrial structure are considered to be the main factors driving the growth of carbon emissions (Li and Wu 2019). It is clearly known that population is always an indispensable factor in carbon emissions related research. Given that, monitoring and analyzing the impact of human factors such as population on carbon emissions has become an important research topic (Zhu et al. 2010).

Highlights:
• Analyze the transmission path and dual transmission mechanism of population factors affecting carbon emissions from the perspective of population flow.
• Comprehensively discuss the effects of different types of population structure changes and other factors on carbon emissions caused by population flow.
• Hu Line is used to examine the impact of regional heterogeneity on the carbon emission effect of population flow. Most literatures related to the impact of population factors on carbon emissions only focus on the impact of total population on carbon emissions. Simultaneously, most scholars believe that the expansion of population size will increase carbon emissions (Albrecht et al. 2002). With the development of society and the differentiation of the internal structure of the population, many scholars have begun to pay attention to the impact of factors such as population urbanization, family size, and age structure of population on carbon emissions (Tian et al. 2015a). Furthermore, the United National Population Fund (UNFPA) pointed out that greenhouse gas emissions are inherently related to factors, such as population growth rate, family size, age composition, urban-rural population ratio, gender, and geographic distribution of population and per capita income.
Population flow is an important factor affecting the change of population structure and economic growth (Duan 2008). Firstly, population flow can change the age structure of population in various regions (Liu 2017). Given that the floating population is dominated by the age-appropriate labor force, population flow will lead to aging acceleration in population outflow areas while aging delay in population inflow areas. Secondly, population flow will promote rural residents to move into cities and towns, further gathering in central cities, which results in urban and rural demographic structure changes as urbanization (Gao and Zhang 2016). Thirdly, population flow will change the family structure; family size tends to be "small" and "simplified" (Shao and Wu 2018). Fourthly, population flow has also changed the knowledge structure of the population in different regions. High educated people generally tend to move to central cities and high-income areas. The larger the city, the more significant net inflow of population (Ding et al. 2018).
Changes in the demographic structure have important impact on regional carbon emissions. Relevant studies have shown that aging has a significant impact on carbon emissions (Dalton et al. 2008;Menz and Welsch 2010) and meanwhile have non-static and incomplete linear relationship (Li 2015). Furthermore, the increase in labor force caused by urbanization and the carbon emissions generated by infrastructure construction cannot be ignored. They are not only closely related (Chen et al. 2020;Yuan and Sun 2020) but also showing a long-term nonlinear relationship (Sun and Huang 2020). At the same time, the change of family size affects household energy consumption to a large extent carbon emission (Cui et al. 2020). The family size effect positively promotes the increase of carbon emissions, while the miniaturization and simplification of family size lead to the decrease of carbon emissions in household consumption. Finally, the change of population knowledge structure is reflected in different education levels. With the increase of education level, the awareness of carbon emission reduction is also increased, and it would promote different energy consumption products. Therefore, the population with higher education level has a significant inhibitory effect on carbon emissions in the future (Yang and Lu 2019). When the overall education level of residents increases, it will inhibit the increase of carbon emissions (Tong et al. 2018).
With in-depth analysis of existing research, it can be found that, though research on the relationship between population and carbon emissions has achieved abundant results, there is still more to explore. First of all, scholars have not conducted in-depth studies on the impact of population flow on carbon emissions, while the research topic should not be ignored. Taking China as an example, its floating population has maintained a large scale years. For example, the floating population in mainland of China was 241 million in 2018, and it remained stable at more than 200 million in the following years. Obviously, this large floating population is an extremely active labor factor, which has a significant role in promoting regional economic growth (Hou and Chen 2016). It is well known that the region with rapid economic growth is also with concentrated population inflows (Yang and Zeng 2014). Population flow has a huge impact on carbon emissions. Secondly, in existing research, most scholars mainly discuss the impact of a single type of demographic change on carbon emissions but rarely consider the comprehensive impact of different types of demographic changes on carbon emissions. Therefore, it is necessary to establish a comprehensive analysis framework for research. Finally, considering China's vast territory, the gaps of economic development, population density, urbanization level, ecological environment, socio-cultural and economic policy in various regions are very large. Regional difference restricts the scale and direction of population flow. Hence, in empirical research, it is necessary to examine the impact of regional heterogeneity on carbon emission resulted by population flow.
The main contributions of this paper are the following: firstly, from the perspective of population flow, this paper analyzes the conduction path and dual conduction mechanism of population factors affecting carbon emissions, which enriches the research between population and carbon emissions. Secondly, it comprehensively explores the impact of different types of population structure changes caused by population flow, as well as other factors on carbon emissions. Finally, since China is divided into two regions named as southeast and northwest with the Hu Line, the impact of the heterogeneity of geographic environment, population density, and economic development on the carbon emission effect of population flow is investigated with it graphically.
The affecting mechanism of population flow on carbon emissions The direct transmission mechanism The most basic economic activity of the human is production and consumption, and energy consumption is the direct source of carbon emissions. Population flow is directly manifested as the shift of population in spatial position. This shift not only manifests as the change of population in different spaces, but also brings about comprehensive adjustment of population structure in space, which includes the changes in population age structure, urban-rural structure, family structure, and knowledge structure. This means that the social and economic activities of the floating population will undergo spatial transfer or quantitative changes during the adjustment process, which will affect the carbon emissions of different regions.
From the perspective of the impact of population flow on production, population inflows and population outflows will affect the size of the regional population (Yu and Kong 2017). It directly leads to spatial changes in the number of labor supply and in turn affects the changes in factor market prices. Meanwhile, according to the change of factor input cost, the production decision of enterprises will be adjusted. The change of production activities leads to the adjustment of energy consumption, which directly affects the carbon emissions in production activities. For example, the development and application of renewable energy in automotive industry can effectively reduce carbon emissions (Kihm and Trommer 2014). In short term, population outflow will lead to a decline in output, while population inflow will lead to an increase in output, thereby affecting the amount of carbon emissions. However, in the long run, due to the adjustment of production factors, decision-making, and the role of technological innovation, the impact of population flow on the output of inflow and outflow areas is uncertain, so as the impact on carbon emissions in production activities.
As for the direct impact of population flow on consumption, population flow makes partial or all consumption transfer spatially, which leads to the growth of consumption in inflow area and the decline of consumption in outflow area. It is acknowledged that regional differences lead to unbalanced spatial economic development. In the field of consumption, this imbalance is manifested in different consumption levels and structures. In essence, there are diversities in the demand structure and quantity of consumer goods and services with different energy densities. Affected by the economic level, social culture, consumption habits, and consumption demand, the inflow population will gradually be integrated into the local consumer groups. Studies have shown that population flow will increase the consumption expenditure of immigrant households (Bălă and Prada 2014) and the heterogeneity within the migrant population causes intergenerational differences in consumption (Wang and Deng 2021). For outflow areas, population outflow also means changes in consumption levels and consumption structure. Therefore, the change of consumption caused by population flow is not a simple transfer in space but accompanied by the time conversion, structural change, preference change, and level adjustment of consumption, which directly affects the change of local energy consumption and carbon emissions.

The indirect transmission mechanism
Population flow also indirectly affects carbon emissions by influencing social, economic, and cultural activities. The main channels are as follows: firstly, it acts on carbon emissions through the impact on economic structure. Population flow is an important factor affecting the regional economic structure (Tang et al. 2017), and it has an impact on the economic structure. The impact includes changes of industrial structure, income structure, savings structure, human capital structure, trade structure, and total factor productivity. These changes will be reflected in the production and consumption of energy quantity and type of demand changes. Moreover, energy consumption has a positive and significant impact on carbon emissions (Apergis and Payne 2009). Secondly, population flow puts forward higher requirements on regional public services, not only requiring multi-dimensional improvement of public services, but also achieving fair access to public services (Ren and Yin 2019). These services provide guarantee for the public to participate in social economic, political, cultural activities, and so on. These guarantees must be at the expense of consuming social public products. Providing such public products will also lead to a large amount of energy consumption, which will lead to an increase in regional carbon emissions. Thirdly, it acts on carbon emissions through its impact on society and culture. Social culture has regional differences, which determine the population's willingness and preference for social and economic activities. The changes in the population structure caused by population flow will accelerate the cultural integration. Changes in different regions change the willingness and preferences of local residents' social and economic activities. However, the residents' personal preferences and behaviors are one of the important factors of carbon emissions (Rong et al. 2020). It leads to changes in the structure and total amount of demand for consumer goods and services with different energy densities and indirectly affects energy consumption and carbon emissions (Iraganaboina and Eluru 2021).
Based on the above analysis and the effect of population flow, there is a common dual transmission mechanism of the impact of population flow on carbon emissions, which includes direct and indirect transmission mechanism, as shown in Figure 1.

Model setting
In order to investigate the impact of regional population flow on carbon emissions, this paper first constructs the following function: LnCE represents regional carbon emissions, and M represents various factors affecting carbon emissions. The function describes the relationship between carbon emissions and its influencing factors. Due to the uncertainty in the economic model, it is necessary to construct a specific econometric model. In order to investigate the impact of population flow on carbon emissions at provincial level in China, this research selects 30 provinces in China, and all sample areas have been included as much as possible. In order to estimate the individual heterogeneity error and time heterogeneity error as much as possible, this paper uses a fixed effects analysis model to improve results consistency. Since the intercepts of cross sections and series are different, models will be different. The time-point individual fixed effects model is constructed. What is more, this paper uses population flow as a core explanatory variable. The basic empirical model is set as follows: Among them, i represents a province; t represents time; LnCE i, t represents carbon emissions; PF i, trepresents current population flow; X represents a control variable that affects carbon emissions; β represents the effect of an explanatory variable on the explained variable coefficient; μ i, t represents time fixed effect; ν i, t represents individual fixed effect; and ε i, t represents random interference term.
Considering that the production and consumption activities of the floating population may lag behind when move to a new place, this paper introduces the one-stage variable PF i, t on the basis of model (2), and model (3) is constructed as follows: Considering that there may be a U-shaped curve relationship between population flow and carbon emissions, on the basis of model (2), the quadratic term of population flow is introduced to construct model (4) as follows: Further considering that population flow will lead to changes in the age structure, urban-rural structure, family structure, and knowledge structure of the population, thus affecting carbon emissions, this paper introduces the cross terms of population flow and aging, urbanization, household size, and education level into the basic empirical model and obtains the specific models (5)-(8) as follows: Among them, and PF i, t PK i, t denote the cross items of population flow and aging, urbanization, household size, and education level, respectively, analyzing the impact of population flow on China's provincial carbon emissions through different transmission paths.
At the same time, in order to examine whether the carbon emissions effects of geographical environment, population density and economic development level on population flow are heterogeneous, this paper applies Hu Line to divide China into southeast and northwest areas. From the perspective of geographical characteristics, southeast areas are mainly plain and hilly terrain with dense water network, which is suitable for farming. While northwest areas are mainly snow plateau and desert area, which is suitable for grazing. From the perspective of population density and economic development level, the southeast areas accounts for 43.18% of the national land area, gathering 93.77% of the country's population and 95.70% of GDP, while the population and economic density of northwest areas are fairly low. From the perspective of urbanization level, it is higher than the national average level in southeast areas, while it is lower than the national average level in northwest areas. Hu Line is not only the dividing line of China's geographical and ecological environment but also the dividing line of population concentration and economic development level. Therefore, this paper uses the Hu Line to divide China's provinces into Hu Line southeast provinces and Hu Line northwest provinces to construct models (9) and (10) to conduct group regression of population flow and control variables, respectively.
Among them, wn represents northwest provinces; es represents southeast provinces; t represents time; LnCE represents carbon emissions; PF represents current population flow; X represents control variables affecting carbon emissions; β represents coefficients of explanatory variables to explained variables; μ represents time fixed effect; ν represents individual fixed effect; and ε represents random interference term.
Variable selection, data source, and processing Variable selection I. Explained variable The explained variable in this paper is the annual carbon emissions of China's provinces. According to Du Limin's calculation method of carbon emissions (Du 2010), this paper selects seven kinds of energy, including coal, coke, gasoline, kerosene, diesel, fuel oil, and natural gas, multiplied the corresponding energy consumption and carbon dioxide factor, and summed to get the carbon emission data of each province. Due to the huge amount of carbon emissions in each region, there is an order of magnitude gap with other variables. We take the log of it.

II. Core explanatory variables
(1) Population flow (PF). The description of floating population generally depends on the statistical data of floating population. At present, there is no unified definition of floating population in academic circles. This paper defines the population flow as the difference value between permanent residents and the registered residents (Shi 2020).
(2) Aging (PA). The aging of population is the most important feature of modern population transformation. At present, in most studies, aging is measured by the proportion of the elderly aged 65 and above in the total population (Li 2019). This paper also uses this index for reference.
(3) Urbanization (PC). The obvious change of urban and rural population structure is reflected in urbanization.
The measurement methods of urbanization generally include the proportion of population, proportion of urban land, and coefficient adjustment method. Considering the availability of data, in this paper, urbanization is measured by the ratio of urban population to total population (Li 2015). (4) Household size (PH). The change of family structure caused by population flow is reflected in family size. In this paper, household size is measured by the average number of each family. (5) Knowledge structure of the population (PK). Some studies pointed out that education has a great dynamic effect on population flow (Meng 1993). The population flow of school-age youth due to receiving education will also affect the knowledge structure of the population in the corresponding region. In this paper, knowledge structure of the population is measured by the proportion of junior college or above.
III. Control variables Based on the research of other scholars, this paper selects five control variables: resident consumption (RC), per capita GDP (PGDP), energy intensity (EI), energy consumption structure (EC), and carbon emission intensity (CI). Residential consumption is measured by the ratio of per capita consumption of urban residents to per capita GDP. Per capita GDP is measured by the ratio of regional gross domestic product (GDP) to permanent resident population. GDP is converted to the constant price level based on 2005 by using the deflator index and then took the logarithm for eliminating the effect of heteroscedasticity; energy intensity is measured by energy consumption per unit GDP; energy consumption structure is measured by the ratio of coal consumption to total energy consumption; the carbon emission intensity is measured by the carbon emission per unit GDP.

Data sources and descriptive statistics
The data is from 2005 to 2018 of China. Since part of the energy consumption data of Tibet cannot be obtained, the regions include 30 provinces except Tibet, Hong Kong, Macao, and Taiwan. The data in 2018 of energy consumption is obtained by linear fitting, and other missing data are supplemented by interpolation method. The data mainly come from multiple bases, including the China Environmental Yearbook, the China Environmental Statistical Yearbook, the China energy statistical yearbook, the China Statistical Yearbook, the China Demographic Yearbook, EPS database, and official website of National Bureau of statistics. The data units and descriptive statistical variables are listed in Table 1.

Data verification
In order to avoid spurious regression, it is necessary to put unit root test and co-integration test on panel data. In this paper, ADF test, LLC test, IPS test, and PP test are used to test the unit root of panel data. The specific test results are listed in Table 2. Although some variables in the IPS test did not pass the significance test, combined with the other three tests, it can be considered that the variables in the previous model are stable.
In this paper, Kao test, Pedroni test, and Westerlund test are used to test the co-integration on variables in each model, so as to explain whether the regression of each model is appropriate. The specific test results are listed in Table 3. It can be found that there is a co-integration relationship between the variables of each model, which indicates that the model setting is appropriate.

Results and analysis
Regression results and analysis of basic model After unit root test and co-integration test based on panel data, Hausman test is carried out to select random effect or fixed effect. The original hypothesis of Hausman test is random effect model. However, according to the test results in Table 4, the null hypothesis was rejected by all the three models at the 5% significance level, so it is reasonable to be explained by fixed effect model. The regression results of basic empirical models are listed in Table 4.
In model (2), the coefficient of population flow is negative and passes the test at the 1% significance level, which indicates that in the short term, population flow reduces carbon emissions in China. In model (3), the coefficient of population flow lagging for one period is still negative at the 1% significance level, indicating that even though the production and consumption activities lag due to population flow, it still has an inhibitory effect on carbon emissions. In model (4), the coefficient of quadratic term of population flow also negatively passes the test at 1% significance level. It indicates that in the long run, population flow in China also reduces carbon emissions. The above elastic coefficients are all small, but considering the huge population base and floating population in China, the impact of population flow on the carbon emissions cannot be ignored. In conclusion, with regression results, we can find population flow in China is beneficial to reducing carbon emissions, but the specific reasons need further analysis.
In the control variables, residents' consumption, per capita GDP, energy intensity, and energy consumption structure have positive impact on carbon emissions in three models. And they all surpass 1% significance level. It indicates that the above factors have a significant driving effect on carbon emissions, which is consistent with the actual situation in China. In recent years, China's economy continues to grow and people's living standards continue to improve, which is directly reflected in the growth of per capita GDP and the increase of household consumption. Therefore, growth of production and consumption lead to the increase of carbon emissions. Energy intensity reflects the economic efficiency of energy. The higher the energy intensity, the higher the energy consumption per unit output; therefore, the increase of energy intensity will also promote the increase of carbon emissions. Although in recent years, China has enhanced energy conservation and emission reduction, it is still difficult to shift to a green growth model due to the large proportion of energyintensive, high polluting industries and the relatively backward technology of energy conservation and emission reduction.
In addition, the proportion of coal in China's energy consumption structure is high, and the environmental pollution is serious, which is also the main reason for increasing carbon emissions. Now, China is accelerating the adjustment of energy consumption structure, promoting the development of green economy, gradually replacing traditional energy with clean energy, reducing the consumption of coal, increasing the supply of natural gas, vigorously developing hydropower resources, and promoting the use of renewable energy. This work improves energy efficiency and significantly reduces carbon emissions.
In control variables, the coefficient of carbon emission intensity on carbon emissions is negative in the three models and they surpass the significance level of 1% or 5%. It indicates that the lower the carbon emission intensity is, the more the carbon emission quantity increases. This is not consistent with common sense. The decrease of carbon emission intensity means that the increase of unit GDP produces less carbon dioxide emissions (Yin et al. 2020). Since carbon emission intensity is the carbon emission per unit of GDP, considering the rapid growth of China's economy in recent decades, this empirical result seems to indicate that the growth rate of China's economy is higher than the decline rate of carbon

Analysis of population structure change's impact on carbon emissions in population flow
In order to further explore the impact of various demographic changes on carbon emissions caused by population flow, this paper introduces the cross terms of population flow and aging, urbanization, household size, and knowledge structure into the basic empirical model, and the specific regression results are shown as Table 5. As summarized in Table 5, that under different population structures, the coefficients of the control variables including residents consumption, per capita GDP, energy intensity, and energy consumption structure are all positive and pass the test at the 1% significance level on carbon emissions. And the coefficients of carbon emission intensity are negative and pass the test at least 5% significance level on carbon emissions. The results are consistent with the basic empirical model, which indicates that the regression results of control variables are highly robust.
In model (5), the coefficient of population flow's impact on carbon emissions is negative, and the cross term of population flow and aging is also negative. The two surpass significance level 1% and 10%, respectively, which indicates that the aging caused by population flow inhibits the increase of carbon emissions in China. The results are consistent with the characteristics of population aging in China. In general, population aging process is accelerating in China, showing the characteristics of large amount, rapid growth, regional imbalance, and aging before getting rich (Li and Zhang 2017), while the population flow further aggravates the spatial imbalance of population aging. On one hand, from the perspective of production, the elderly people in China seldom continue to work after retirement. On the other hand, in the perspective of consumption, because China has entered the aging society when the overall economy is still underdeveloped, the income level of the elderly is generally not high. Current old-age medical security system, old-age care service system, and socialized management system are not able to meet the growing demand for the elderly. Therefore, population aging not only reduces the carbon emissions in the production field, but also leads to the shift of consumption structure to low carbon and energy saving, thus inhibiting the carbon emissions on the whole.
In model (6), the coefficient of the cross term between population flow and urbanization is positive, but it does not pass the significance test. This indicates that there is no significant correlation between the increase of urbanization level caused by population flow and the growth of carbon Note: ***, **, and * denote statistical significance levels at 1%, 5%, and 10%, respectively Note: ***, **, and * denote statistical significance levels at 1%, 5%, and 10%, respectively emissions. The results are consistent with the characteristics of urbanization in China. Since the reform and opening up, population urbanization has rapidly reached 60% by 2018 in China, but the quality is not high. Population urbanization lags behind land urbanization. The reason is that Chinese urbanization is mainly manifested as urbanization of residence registration rather than real residence, that is, urbanization of population at the economic, social, and cultural psychological level. Therefore, the improvement of China's urbanization level does not really reflect the role or status of population flow in the field of production and consumption. There is no significant correlation between urbanization level and the growth of carbon emissions. In model (7), the cross terms of population flow and household size did not pass the significance test. This result indicates that there is no significant correlation between carbon emissions and household miniaturization caused by population flow. Generally speaking, due to the scale effect of household consumption, the total energy demand of small households will exceed that of large households. Therefore, household miniaturization will increase carbon emissions. However, in China, population flow not only makes the family size smaller but also makes the internal structure of the family simpler, as well as the proportion of single-person households and generational households increase (Zhou 2016). Most of these families have simple life, low commodity, and energy consumption. Therefore, there is no significant correlation between household miniaturization caused by population flow and carbon emissions.
In model (8), the cross term of population flow and population knowledge structure is negative, which pass the test at the 1% significance level, indicating that the improvement of population knowledge structure caused by population flow in China inhibits carbon emissions in China. According to the data of "China floating population development report (2016)," in recent years, the floating population with higher education is increasing. The proportion of the floating population with the purpose of development and learning and training is increasing, especially in the new generation of young floating population. Highly educated population is more likely to accept the idea of energy conservation and emission reduction and put it into action, thus reducing carbon emissions.

Heterogeneity regression based on Hu Line
Geographical location, population density, and level of economic development all limit or affect population flow and then affect the carbon emissions of different regions. Table 6 shows the results of heterogeneity regression based on the Hu Line.
Except for the insignificant regression results of individual variables in models (9) and (10), most of the control variables have highly similar impact on carbon emissions as the basic model, and they all surpass significance level 1% without obvious regional heterogeneity, which indicates that the regression results of control variables are still highly robust.
Comparing the results of model (9) and (10), the impact coefficient of population flow on carbon emissions is positive in the group of provinces located in the northwest of Hu Line, while it is negative in the group of provinces located in the southeast of Hu Line, and all of them surpass 1% significance level.
The result shows that, in the region with poor ecological environment, sparse population, and relatively backward economic development, population flow leads to the increase of carbon emissions; in the region with good ecological environment, suitable climate, convenient transportation, and developed economy, population flow leads to the decrease of carbon emissions. The reason is obvious. In order to overcome the vast and harsh natural environment, people in northwest areas of China need to consume more energy in infrastructure construction, transportation security, and life quality improvement. And the inflow of population will increase carbon emissions to a large extent. On the contrary, in southeast areas, the climate is suitable for living, and production, the dense population, and developed economy make the scale effect and intensive effect exist simultaneously and improve the energy utilization efficiency, thus restraining the growth of carbon emissions. When observed the direction of population flow in China, eastward and southward migration and concentration to core cities are the main characteristics. In 2018, 7 of the 10 provinces in eastern China are net population inflow areas, and the net inflow population in southern provinces was 1.685 million. This phenomenon fully demonstrates that geographical and economic factors play an important role in the population flow. At the same time, combined with the empirical results of model (2), it can be inferred that even if there is an increase in carbon emissions caused by population inflows in the northwest of China, but due to the population mainly Note: ***, **, and * denote statistical significance levels at 1%, 5%, and 10%, respectively inflows to the southeast, the overall carbon emissions is still reduced for population flow.

Discussion
This study is based on the premise that the main movement of population flow is due to economic factors, and the empirical results are based on China's population flow data. However, the factors driving population flow are not all economic factors, and other factors such as politics, religion, and war may be the main factors leading to population flow. The analytical framework established in this paper does not apply to the analysis of the carbon emission effect of such population flow.
In fact, the growth of carbon emissions is mainly related to economic development, and the main motivation of China's population flow is economic driving. From this point, the conclusion of this paper is universal. Population flow is a complex social phenomenon, which includes not only inter-provincial flow but also inter-city, suburban, and urban flow within the province. Studies have shown that inter-provincial population flow mainly depends on the level of economic development and manufacturing employment opportunities in the inflow region, while intra-provincial population flow mainly depends on the level of public services in the inflow region (Tian et al. 2015b). Therefore, the impact of inter-provincial population flow and intra-provincial population flow on carbon emissions is significantly different. However, due to the availability of data, this study does not analyze population flow in the province, so the empirical results of this paper cannot reflect this difference.
In the meantime, population flow and its resulting changes in population structure have a strong spatial spillover effect, which leads to the spatial correlation and dependence of regional carbon emissions. Although there are many studies on the spatial spillover of carbon emissions (Tao et al. 2016), there are few studies on the spillover effect of carbon emissions from the perspective of population flow. More studies are limited to the population, population growth, population proportion, and human capital as the influencing factors of carbon emissions (Tong et al. 2015;Song 2017). But some scholars try to analyze the spatial spillover effect of carbon emissions directly from the perspective of population structure change. For example, the research of Cheng et al. (2019) shows that heterogeneous population urbanization has different spatial spillover effects on carbon emissions. Therefore, the next research should incorporate the demographic changes caused by population flow into the spillover analysis framework and analyze the relationship between population flow and its structural changes and spatial spillover of carbon emissions.

Conclusion
The population flow can directly bring about the change of regional population structure and lead to the spatial transfer of production and consumption, thus affecting the growth of regional carbon emissions. The change of population structure will also indirectly affect the growth of regional carbon emissions through the adjustment of social, economic, and cultural activities. Based on the analysis of the dual transmission mechanism of the effect of population flow on carbon emissions through population structure, this paper empirically analyzes the impact of population flow and other related factors on carbon emission with panel econometric model and heterogeneity analysis.
The main finding reveals that, firstly, for China as a whole, whether in the long run or in the short run, population flow can reduce the growth of carbon emissions. Even considering the impact of production and consumption lag caused by population flow, population flow also has a positive effect on carbon emission reduction. Secondly, the aging and knowledge structure improvement of population caused by population flow helps to reduce carbon emissions, but the urbanization and household miniaturization caused by population flow have no significant correlation with the growth of carbon emissions. Thirdly, the results of group regression based on Hu Line show that the heterogeneity of geographical environment, population density, and economic development has a certain impact on the carbon emission effect of population flow. In areas with good ecological environment, dense population, and developed economy, population flow reduces carbon emissions, while in areas with poor ecological environment, sparse population, and backward economy, population flow increases carbon emissions. Fourthly, level of residents consumption, per capita GDP, energy intensity, and energy consumption structure in China have positive effects on carbon emissions, while carbon emission intensity has negative effects on carbon emissions. Due to the continuous growth of China's economy, increase of per capita GDP and energy intensity become the main factors to promote carbon emissions.
However, there are still some limitations in this paper; the most important one is that it does not consider the spatial correlation and dependence caused by population flow among regions. In fact, China's population flow and its impact present a spatial spillover effect. Incorporating spatial factors into the empirical test of the mitigation effect of population flow is an important direction for future research. In addition, population flow includes not only inter-provincial mobility but also intra-provincial mobility. However, due to the unavailability of data, this paper did not analyze the population flow in the province.

Policy implications
Based on the major findings from this paper, to coordinate population policy with energy conservation and emission reduction policy, the following policy implications are drawn: Firstly, the government needs to improve the management, service, and integration of the floating population. Through scientific and efficient management, the population flow tends to areas with good environment and developed economy. Strengthen the supply of basic public services with fairness and efficient, thus to realize the full coverage of regional population and reduce the waste of resources caused by the lack of public services. Deepen the reform of urban management system and social policy; make full use of modern technical; achieve dynamic, convenient, sustainable, and intelligent service development mode, which could not only promotes the harmonious integration of the floating population, but also effectively reduces the service cost and carbon emissions.
Secondly, the aging service system needs to be improved, and aging industry needs to be developed vigorously with the goal of low-carbon environmental protection and green health. It is necessary to actively implement the 2019 "China's medium and long-term plan for actively coping with population aging" to improve the elderly care and health service system, as well as to establish a comfortable social environment for the elderly. It is necessary to accelerate the development of the elderly service industry with low-carbon environmental protection and green health as the main goal to drive the development of low-carbon economy combining the characteristics of the elderly industry such as high comprehensiveness, long industrial chain, high relevance, and wide range of fields. Furthermore, it is essential to allocate labor resources reasonably, to guide labor flow to the elderly service industry and to provide diversified services for the elderly.
Thirdly, to promote the development of low-carbon economy, it is necessary to strengthen the interaction between population structure and industrial structure. On one hand, it is necessary to accelerate the reform of education and training system, upgrade population knowledge, and its structure to provide a large amount of high-quality human capital for the upgrading of industrial structure. On the other hand, optimizing the industrial structure contributes to accelerating the transformation of high-tech industries. Enhance the development of modern service industries with low energy consumption, such as eco-cultural tourism, modern finance, science and technology services, information services and ecommerce that is beneficial, thus to meet the needs of highquality development of population.
Fourthly, accelerate the new urbanization and achieve the goal realizing the social and economic role transformation of the transferred population. For urbanization, it is essential to promote the reform of household registration, rural land property rights, and social security system and to solve problems of employment, education, housing, pension, and education. What is more, it is necessary to realize not only the transformation of population, identity, and occupation but also the transformation of thinking, knowledge, and behavior mode to adapt to the development mode of low energy consumption, low emission, and low pollution in modern cities and towns.