Population shrinkage in China’s three major urban agglomerations: spatiotemporal patterns and driving factors

1 Introduction

Urban population shrinkage has emerged as a global urban challenge, manifesting as a multifaceted dilemma featuring continuous population outflow, declining economic vitality, and socio-spatial deterioration. Since the mid-20th century, traditional industrial regions such as Germany's Ruhr region and the U.S. "Rust Belt" have been the first to undergo this process, prompting systematic academic research on "shrinking cities" (Beauregard, 2009; Martinez-Fernandez et al., 2012). As China's urbanization enters its mid-to-late transformation phase, population shrinkage has gradually emerged and is spreading, particularly in the Northeast region, resource-based cities, and the peripheral areas of certain coastal urban agglomerations. China is now at a critical juncture of demographic transition: slowing total population growth and intensified spatial redistribution have created a complex pattern within major urban agglomerations, characterized by the coexistence of population agglomeration and local shrinkage (He et al., 2023; Zheng and Zhang, 2025; Liu and Zhu, 2023; Li et al., 2025; Jin and Sui, 2021; Yang and Dunford, 2018). This phenomenon poses severe challenges to urban sustainable development, regional economic resilience, and social governance, necessitating a systematic understanding of its spatiotemporal evolution patterns and underlying mechanisms.

Understanding the mechanisms of population mobility and shrinkage in cities of different tiers and types holds crucial research value for forecasting regional population distribution patterns and understanding the interaction between policy interventions and market forces. This policy relevance is underscored by China's national urban development strategy, which explicitly calls for orderly relocation of non-core functions from megacities and dynamic optimization of infrastructure and public services in line with permanent resident population trends. This policy orientation reveals the direction of structural adjustments in China's population mobility and urban development patterns for the foreseeable future, yet the academic understanding of how these policy interventions interact with market-driven population flows remains incomplete.

In terms of academic progress, research on population shrinkage, despite substantial progress, has expanded from simple demographic descriptions to multidimensional, comprehensive measurements, multiscale spatial analysis, and in-depth exploration of influencing factors and governance responses. However, existing research still has the following shortcomings: first, studies have primarily focused on individual cities or specific types of cities, with relatively limited comparative research on cross-regional, multi-type urban agglomerations; second, although existing studies have noted the coexistence of growth and shrinkage within urban agglomerations, there is a lack of systematic comparisons of the spatiotemporal evolution of shrinkage across the three major urban agglomerations—Beijing-Tianjin-Hebei, the Yangtze River Delta, and the mainland cities of the Guangdong-Hong Kong-Macao—as well as quantitative analyses of their driving mechanisms; third, analyses of influencing factors have largely concentrated on economic and demographic variables, with insufficient comprehensive consideration of dimensions such as the ecological environment, social capital, and policy interventions;Fourth, the exploration of multiscale correlations and spatial spillover effects in shrinkage research remains in its infancy.

Based on the above theoretical and practical background, this study focuses on the Beijing-Tianjin-Hebei, Yangtze River Delta, and the mainland cities of the GBA. They represent China's most economically dynamic and densely populated regions, corresponding to three distinct development models—northern, eastern coastal, and southern coastal—which together capture the diversity and complexity of urban population changes across different regional contexts. Moreover, the growing divergence in population growth within these agglomerations in recent years offers timely insights into regional coordination challenges and provides forward-looking references for other urban agglomerations. Additionally, the relatively well-established statistical systems in these regions ensure data availability and reliability for the empirical analysis. Using panel data from 2010 to 2024, the study first identifies the spatiotemporal evolution characteristics of population shrinkage in each urban agglomeration, and then constructs a two-way fixed effects panel regression model, to systematically examine the impact of factors such as natural population growth, economic development, industrial structure, transportation infrastructure, public services, and the ecological environment on population changes, and to reveal regional heterogeneity through between-group coefficient tests. The study found that: the BTH urban agglomeration exhibits a gradual shrinkage path characterized by “peripheral spread—core retention,” with Beijing and Tianjin falling into the mild shrinkage category under the dual effects of policy-driven relocation and structural adjustment; the YRD urban agglomeration shows an evolutionary pattern of “wide diffusion of mild shrinkage and localized persistence of severe shrinkage”; Shanghai has experienced mild shrinkage for the first time, but the shrinkage intensity in most cities previously classified as severely contracting has eased; the mainland cities of the GBA, however, has consistently had no contracting cities, demonstrating strong resilience in population aggregation. Regarding influencing factors, population changes in the BTH region were significantly and negatively affected only by passenger traffic volume, with a marked outflow of peripheral populations due to the siphoning effect of core cities; the YRD was driven by multiple factors, with natural population growth rate, per capita GDP, and road area showing significant positive effects, while tertiary industry GDP, number of high-speed rail stations, and passenger volume had a significantly negative effect; in the mainland cities of the GBA, natural population growth rate and the number of high-speed rail stations served as core positive driving factors. Inter-group coefficient tests further reveal significant regional differences in the effects of influencing factors across the three major urban agglomerations, and endogeneity tests validate the reliability of these conclusions.

This study makes two primary contributions to the existing literature. Theoretically, it is the first to incorporate the three major urban agglomerations into a unified comparative analytical framework, moving beyond the predominant focus on a single urban agglomeration or specific city types in previous research. By systematically comparing the spatiotemporal evolution and driving mechanisms of population shrinkage across regions with distinct development models, this study advances the theoretical understanding of the general patterns and regional specificities of urban shrinkage. Practically, at a time when China is experiencing a demographic turning point and a new phase of urbanization, the coexistence of growth and shrinkage within urban agglomerations has become an unavoidable reality. The findings provide empirical evidence for differentiated urban policy-making and support the transition of urban agglomerations from scale expansion to high-quality development, aligning closely with the "category-based policy" approach in China's New Urbanization Plan (Central Committee of the Communist Party of China & State Council, 2025).

2 Literature review

2.1 Theories of population shrinkage

Population shrinkage, also known as population loss, population reduction, or population decline, originated from the demographic transition and industrial restructuring in Western developed countries during the mid-20th century. With the decline of traditional industries, cities in regions such as the Ruhr region of Germany and the U.S. “Rust Belt” have faced structural challenges of population shrinkage, shrinking employment, and economic recession since the 1950s (Beauregard, 2003; Martinez-Fernandez et al., 2012). German scholars Häußermann and Siebel first introduced the concept of “shrinking cities” in the 1980s to describe cities experiencing sustained population loss, economic decline, and physical deterioration. Since then, population shrinkage has gradually become a frontier topic in global urban studies. According to statistics by Martinez-Fernandez et al. (2016), over the past two decades, more than a quarter of the world’s major cities have experienced varying degrees of population loss. In China, as economic transformation and urbanization deepen, the phenomenon of population shrinkage has become increasingly prominent, particularly in the Northeast, resource-based cities, and the peripheral areas of certain coastal urban agglomerations (Li et al., 2020; Xian and Chen, 2022). Population shrinkage involves not only a reduction in population size but also multidimensional structural changes across economic, social, and spatial dimensions, posing a severe challenge to urban sustainable development (Cai et al., 2026).

2.2 The concept and identification of urban shrinkage

The core characteristic of urban shrinkage is sustained population loss, but its causes and manifestations are complex and diverse. Early studies often used the total population as the sole identification criterion (Hollander and Németh, 2011), but as theoretical understanding deepened, scholars gradually recognized that shrinkage is a multidimensional phenomenon that requires comprehensive measurement across multiple dimensions, including population, economy, society, space, and the environment. For example, Liu et al. identified shrinking cities in China using population and economic data, finding that the Northeast region experienced the most significant shrinkage. Wang et al. constructed an evaluation system across three dimensions—population, economy, and space—and their analysis of 175 resource-based cities indicated that the number of shrinking cities is on the rise, with a spatial distribution characterized by concentration in the Northeast (Liu et al., 2018; Wang et al., 2021). Regarding identification methods, in addition to the traditional annual average change rate of permanent residents, new data sources and methods such as nighttime light data (Liu et al., 2019; Zheng and Zhang, 2025), Baidu migration big data (Meng et al., 2021), and machine learning clustering (Ma and Yoon, 2025) have been widely applied, enhancing the accuracy and timeliness of shrinkage identification. Ma & Yoon employed dynamic time-series clustering to classify 285 Chinese cities into four distinct shrinkage-growth trajectories, revealing heterogeneous evolutionary paths for different types of shrinkage (Ma and Yoon, 2025). Furthermore, multiscale analysis has emerged as a significant trend. Ma et al. found that urban shrinkage shows significant differences across administrative levels (prefecture-level cities, county-level cities, and towns) and that spatial spillover effects occur between these scales (Ma et al., 2023; Ma et al., 2024). Cai et al., based on a systematic review of 183 global literature sources using large language models, further pointed out that shrinking cities possess a unique developmental logic, manifested as a complex evolutionary pattern where local population inflows coexist with overall shrinkage, and the decline of core areas coexists with the expansion of peripheral areas, thereby transcending the traditional binary cognitive framework (Cai et al., 2026).

2.3 Spatial correlations between urban agglomerations and population shrinkage

As the primary form of regional spatial organization, urban agglomerations profoundly influence the spatial patterns of shrinkage and growth through internal population flows and redistribution. In rapidly urbanizing China, while populations are concentrating in coastal urban agglomerations and inland provincial capitals, localized shrinkage phenomena are also spreading (Cai et al., 2026). Guojun et al. analyzed the spatiotemporal characteristics of China’s population distribution from a metropolitan area perspective, noting that the Guangdong-Hong Kong-Macao, Yangtze River Delta, and BTH regions experience both population inflows and outflows, with some cities already showing signs of shrinkage (Guojun et al., 2018). Xian and Chen, based on data from the sixth and seventh national censuses, found that 92.7% of prefecture-level cities in Northeast China experienced population loss, forming a typical shrinking region, while coastal urban agglomerations and inland provincial capitals constitute new population growth poles (Xian and Chen, 2022). Ma et al. conducted a multiscale analysis of Jilin Province, revealing that growth and shrinkage at the prefecture-level city level are driven by economic and political factors, whereas those at the county and town levels are more influenced by natural and social conditions. Furthermore, a scale-dependent relationship exists between higher- and lower-level cities: towns under the jurisdiction of counties experiencing absolute growth may exhibit moderate shrinkage, while towns under counties experiencing absolute shrinkage face a risk of absolute shrinkage (Ma et al., 2023). Ma et al. further emphasize that factors such as traditional industries, surplus agricultural labor, and educational resources influence the population trajectories of cities at different scales through spatial spillover effects; the pattern of population shrinkage is the result of the combined effects of regional economic restructuring, policy interventions, and natural conditions (Ma et al., 2024). Therefore, understanding the spatial mechanisms of population shrinkage at the urban agglomeration scale is of great significance for formulating differentiated regional policies.

2.4 Multidimensional factors influencing population shrinkage

The causes of population shrinkage are highly heterogeneous and scale-dependent, involving multiple factors such as resource depletion, industrial restructuring, locational marginalization, policy shifts, natural population fluctuations, and changes in the ecological environment. Research by Bartholomae and others indicates that deindustrialization, aging, suburbanization, and global competition are the primary drivers of shrinking cities in the West (Bartholomae et al., 2017; Hattori et al., 2017; Khavarian-Garmsir et al., 2019). In the Chinese context, Li et al. analyzed resource-based cities and found that the lack of industrial support following resource depletion, poor adaptation to market-oriented reforms, and environmental degradation are the core causes, with the shrinkage issues being particularly pronounced in resource-depleted, forestry-based, and Northeast Chinese cities (Li et al., 2020). Ma et al. further demonstrated, through their study in the Northeast region, that the growing pains of structural reform, locational marginalization, and a shortage of surplus agricultural labor constitute the key causes of shrinkage at the prefecture-level city, county-level city, and town scales, respectively (Ma et al., 2024). Ma and Yoon employed time-series clustering to classify Chinese cities into four categories, revealing that resource depletion, manufacturing decline, industrial restructuring, and population migration are the core drivers of distinct types of urban shrinkage (Ma and Yoon, 2025). The impact of transportation infrastructure is twofold: Deng et al. found that high-speed rail may exacerbate the shrinkage of some cities (Deng et al., 2019), while Yang et al. noted that high-speed rail helps mitigate population outflow and promote manufacturing upgrading at the regional level, though its effects vary depending on the route’s location and industrial base (Yang et al., 2024). In recent years, ecological and environmental factors have gradually come into the spotlight. An analysis by Deng et al. of 89 shrinking cities nationwide found that the mitigating effect of shrinkage on urban heat island intensity varies significantly across climatic zones (Deng et al., 2026). Ma and Yoon confirmed that shrinking cities with declining populations can alleviate regional warming (Ma and Yoon, 2024). Xiao et al. compared carbon emission patterns between the Northeast and the YRD urban agglomerations, finding that carbon emissions in rapidly shrinking cities continued to rise, while slightly shrinking cities achieved emission reductions but at the cost of a decline in the secondary sector (Xiao et al., 2019). Furthermore, social capital and public services also influence the process of urban shrinkage. Franklin found that population loss in U.S. cities is unevenly distributed across different spatial scales and population subgroups (Franklin, 2021); Cheng et al. revealed significant heterogeneity in social capital within shrinking cities through their study of Hegang (Cheng et al., 2023); Sha and Cheng further noted that disparities in different groups’ ability to access social capital through green spaces in shrinking cities exacerbate health inequalities (Sha and Cheng, 2024). Regarding policy and governance responses, a comparative study by Béal et al. on shrinking cities in France revealed diverse pathways for “moderate-scale” strategies (Béal et al., 2019); Özatagan and Eraydin found that local growth coalitions in shrinking cities adjust their discourses to continue pursuing growth-oriented policies (Özatağan and Eraydin, 2021); Peters et al. proposed the concept of “smart shrinkage,” emphasizing the improvement of quality of life amid population shrinkage through the development of social infrastructure (Peters et al., 2018).

2.5 Literature review and research positioning

In summary, research on population shrinkage has expanded from simple demographic descriptions to multidimensional comprehensive measurements, multiscale spatial analysis, and in-depth exploration of influencing factors and governance responses. However, existing research still has the following shortcomings: first, studies have primarily focused on individual cities or specific types of cities (such as resource-based cities or old industrial cities), with relatively limited comparative research on cross-regional, multi-type urban agglomerations; second, although previous studies have noted the coexistence of growth and shrinkage within urban agglomerations (Xian and Chen, 2022;Guojun et al., 2018), there is a lack of systematic comparison of the spatiotemporal evolution of shrinkage across the three major urban agglomerations and quantitative analysis of their driving mechanisms; third, analyses of influencing factors have largely concentrated on economic and demographic variables, while comprehensive consideration of dimensions such as the ecological environment, social capital, and policy interventions remains insufficient (Chen et al., 2023); and fourth, the exploration of multiscale correlations and spatial spillover effects in shrinkage research remains in its infancy (Ma et al., 2023). A further critical issue concerns the explanatory power of conventional socioeconomic factors. If economic growth and living environment improvements are widely shared across cities, these factors alone cannot adequately explain why some cities experience population decline while others continue to grow (Sun et al., 2026). Indeed, recent studies have questioned the extent to which aggregate-level economic indicators can capture the divergent population trajectories observed within the same regional context. The digital economy has been found to significantly catalyze urban shrinkage by accelerating the outflow of population, technology, and capital, with pronounced regional heterogeneity (Li et al., 2024). From a physical city perspective, shrinking cities in the Chengdu-Chongqing economic circle have been identified, confirming that factors such as industrial structure transformation and locational conditions interact with broader socio-economic processes to shape population outcomes (Cao et al., 2024). Drawing on network externality theory, urban growth and shrinkage in Northeast China exhibit significant divergent characteristics, and inter-city network connectivity—measured through corporate headquarters-branch networks—offers explanatory power beyond city-level economic aggregates alone (Sun et al., 2026). These findings collectively suggest that the relationship between economic development and population change is not monolithic; rather, it is mediated by network structures, technological transformation, and spatial spillovers. The conventional factors included in our model—GDP, industrial structure, and transportation infrastructure—capture only the direct effects of economic development, while the indirect effects operating through inter-city networks and digital transformation may explain why cities at similar development levels experience divergent population trajectories. We acknowledge this limitation and identify the integration of network externality and digital economy perspectives as a promising direction for future research. Based on this, this study focuses on the Beijing-Tianjin-Hebei, Yangtze River Delta, and the mainland cities of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) as research subjects. Using panel data from 2010 to 2024, it first identifies the spatiotemporal evolution characteristics of population shrinkage in each urban agglomeration, and then constructs a two-way fixed effects panel regression model, to systematically examine the impact of factors such as natural population growth, economic development, industrial structure, transportation infrastructure, public services, and the ecological environment on population changes, and to reveal regional heterogeneity through inter-group coefficient tests. This study aims to fill a gap in comparative research on interregional urban agglomeration shrinkage and to provide a scientific basis for the formulation of differentiated urban policies.

3 Methodology (research methods and data)

3.1 Methodology for identifying population shrinkage in the three major urban agglomerations

3.1.1 Study areas and data sources

This study selects the Beijing-Tianjin-Hebei, Yangtze River Delta, and the mainland cities of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) as research subjects, with prefecture-level cities within each agglomeration serving as the basic units of analysis. Specifically, the mainland cities of the GBA in this study refer to nine mainland cities, namely Guangzhou, Shenzhen, Zhuhai, Foshan, Huizhou, Dongguan, Zhongshan, Jiangmen, and Zhaoqing. Due to data unavailability and statistical caliber differences, Hong Kong and Macao are not included in the analytical framework of this study. Therefore, all references to the "Guangdong-Hong Kong-Macao Greater Bay Area" or "GBA" hereafter in this paper denote the aforementioned nine mainland cities, not the entire Guangdong-Hong Kong-Macao Greater Bay Area. The data covered in this study spans the period from 2010 to 2024. All data are sourced from the statistical yearbooks of the provinces involved in each urban agglomeration, the statistical yearbooks of the respective prefecture-level cities, and the Statistical Bulletins on National Economic and Social Development.

3.1.2 Identification of urban shrinkage

Currently, there is no unified standard for urban shrinkage research. This study references the definition provided by the World Network on Shrinking Cities and uses the permanent resident population to identify shrinkage trends within each urban agglomeration. A negative growth rate in the permanent resident population over a specific period is considered an indicator of population shrinkage. The annual average growth rate (R) of the permanent resident population is used as the criterion for determining growth or shrinkage, the calculation method is shown in Equation 1:

R ( t , t + n ) = M t + n M t n − 1 (1)

where R is the annual average growth rate, t is the starting year, n is the interval length in years, and M is the permanent resident population of the region in that year. The mathematical principle of this formula is to solve for an equivalent constant annual growth rate R that satisfies the equation (1 + R)^*n* = M_t_+*n*/M_t, where the total change over the n-year period is evenly distributed to each year through geometric averaging. Therefore, regardless of whether the interval is 5 years or 6 years, the calculated R always represents the annual average rate of change during that period, ensuring comparability across different time intervals. According to the concept of population shrinkage, when R < 0, the city is in a state of population shrinkage, and the greater the absolute value of R, the more severe the shrinkage; when R ≥ 0, the city is not experiencing population shrinkage. In this paper, cities with an annual average population growth rate lower than −0.5% (R < −0.5%) are defined as experiencing severe shrinkage; those with a rate greater than or equal to −0.5% and less than 0 (−0.5% ≤ R < 0) are defined as experiencing mild shrinkage; those with a rate greater than 0 but less than 1% (0 < R < 1%) are defined as experiencing low-speed growth; and those with a rate greater than or equal to 1% (R ≥ 1%) are defined as experiencing high-speed growth.

3.2 Panel regression model of factors influencing population shrinkage in the three major urban agglomerations

3.2.1 Panel regression models

Static panel data models are the basic form of panel regression and comprise two categories: the fixed-effects model and the random-effects model (Li et al., 2024). The fixed effects model assumes that individual heterogeneity is correlated with explanatory variables; it eliminates individual characteristics that do not change over time through within-group transformations and is suitable for inferring characteristics limited to individuals within the sample. The random-effects model, on the other hand, assumes that individual heterogeneity is uncorrelated with the explanatory variables; it employs generalized least squares estimation and is suitable for inferring population characteristics from the sample. The Hausman test is commonly used to screen between and within models and to test model specifications. The Oxford Handbook of Panel Data, edited by Baltagi, systematically reviews the applications of static panel models in various economic issues, covering classic topics such as production function estimation, labor market analysis, and international trade gravity models. In population and urban studies, static panel models are widely used to identify the drivers of urban population change. For example, Chou (2013) employed a panel data model to analyze the impact of tourism development on economic growth in transition economies, finding that tourism has a significant positive effect on per capita GDP, and that this effect remains robust even after controlling for individual heterogeneity. Jauch and Watzka (2016) employed panel data to analyze the relationship between financial development and income inequality, revealing differentiated impacts of financial deepening across countries at different stages of development (Thayasivam et al., 2025; Lee and Yu, 2024.). These studies provide a methodological foundation for understanding the socioeconomic drivers of urban population changes.

3.2.2 Research design and data description

• (1) Research Subjects and Data Sources

This study selects the mainland cities of the GBA, the YRD urban agglomeration, and the BTH urban agglomeration as the research sample to construct panel data for empirical analysis. Specifically, the mainland cities of the GBA comprise 9 cities and 135 observations; the YRD urban agglomeration comprises 27 cities and 404 observations; and the BTH urban agglomeration comprises 14 cities and 208 observations.

• (2) Variable Definitions

Dependent Variable: Annual Growth Rate of Permanent Resident Population (y), reflecting the intensity of population change in each city.

Explanatory Variables: Covering six dimensions, including population growth, urbanization, economic level, industrial structure, transportation infrastructure, and public services. These dimensions are operationalized through the following specific indicators: natural growth rate (x1), urbanization rate (x2), GDP (x3, in billions of yuan), per capita GDP (x4, in ten thousand yuan), gross output of the secondary sector (x5, in billions of yuan), gross output of the tertiary sector (x6, in billions of yuan), number of completed high-speed rail stations (x7), passenger volume (x8, in ten thousand passenger trips), green space area (x9, in hectares), road area (x10, in ten thousand square meters), number of full-time teachers (x11, in ten thousand), healthcare professionals (x12, in persons), total length of highways in operation (x13, in kilometers), and current-year investment in municipal public utility construction (x14, in ten thousand yuan).

• (3) Model Specification

A two-way fixed-effects panel data model is used to simultaneously control for the effects of individual heterogeneity and time trends on the population change.The basic form of the model is specified in Equation 2:

y it = M it − M i , t − 1 M i , t − 1 × 100 % (2)

Where “y_it” is the annual growth rate of the permanent resident population of city i in year t; “M_it” and “M i , t − 1” denote the permanent resident population of city i in year t and year t−1,respectively. A negative “y_it” indicates population shrinkage, while a positive value indicates population growth. City-specific and year-specific fixed effects are included to absorb unobserved time-invariant city characteristics and common temporal shocks, respectively.

4 Analysis of population shrinkage in the three major urban agglomerations

4.1 Population shrinkage in the BTH urban agglomeration

To ensure an accurate representation of population decline within each urban agglomeration, we defined the study period based on data availability and analytical relevance. Specifically, for the Beijing-Tianjin-Hebei (BTH) urban agglomeration, we quantified population shrinkage using annual growth rate data spanning from 2010 to 2024, as presented in Tables 1, 2 and illustrated in Figure 1.

Table 1. Cities experiencing population shrinkage in the BTH urban agglomeration.

Table 2. Number and proportion of cities with population shrinkage in the BTH urban agglomeration.

Between 2010 and 2015, population shrinkage within the BTH urban agglomeration was observed only in isolated cases, while low-speed growth prevailed as the overarching trend, with core cities maintaining relatively high growth rates. During this period, Dingzhou was the sole municipality to experience mild shrinkage, accounting for 6.25% of the total cities examined (It should be noted that Dingzhou and Xinji are county-level cities, whereas the remaining cities are prefecture-level; owing to discrepancies in administrative boundaries, they cannot be represented on a single map). Geographically situated between Shijiazhuang and Baoding in the central Hebei Plain, Dingzhou benefits from favorable transportation links. Nevertheless, under the strong gravitational pull of the two core metropolitan centers—Beijing and Tianjin—the locality has experienced a modest net population outflow. Critically, no city in the agglomeration has experienced severe population decline, suggesting that the region's overall demographic attractiveness has remained relatively stable.

As the coordinated development strategy for the BTH region was progressively implemented, the period from 2015 to 2020 witnessed a sharp expansion in both the magnitude and spatial scope of population decline. Geographically, this trend exhibited a pronounced diffusion pattern along the northwest–southeast axis. The number of cities experiencing mild population shrinkage rose to four—Baoding, Cangzhou, Handan, and Tangshan—accounting for 25% of the total. In spatial terms, these four cities are distributed precisely along the periphery of the two core metropolitan centers, Beijing and Tianjin: Baoding lies to the southwest of Beijing, Cangzhou to the southeast of Tianjin, Handan in southern Hebei, and Tangshan in eastern Hebei, collectively forming a ring of mildly shrinking cities encircling the two cores. Meanwhile, the number of severely contracting cities surged to 7, accounting for 43.75% of the total. Geographically, these are primarily concentrated in three subregions. The first is the northwestern ecological barrier zone, comprising Zhangjiakou and Chengde, which are situated along the Yanshan–Taihang mountain ranges and assume a critical role in ecological conservation. The second encompasses the traditional agricultural and industrial areas of central and southern Hebei, including Xingtai, Hengshui, and Xinji. The third is Tianjin, located on the eastern coast. Notably, Tianjin—a municipality directly under the central government—experienced its first-ever recorded population decline during this period, signaling that even core cities are beginning to confront mounting pressures on population growth.

Between 2020 and 2024, population shrinkage in the BTH region became a region-wide phenomenon, as core cities joined the contracting group and the spatial extent of decline expanded from the periphery to encompass the entire area. The scope of mild population shrinkage expanded rapidly to nine cities, accounting for 56.25% of the total, and covering all directional quadrants of the BTH region: Chengde and Qinhuangdao to the north; Beijing, Langfang, and Dingzhou in the central belt; Tianjin to the east; Cangzhou and Hengshui to the south; and Anyang in Henan Province. Notably, both Beijing and Tianjin entered the mild shrinkage category, marking the first instance of negative population growth in these two core metropolitan centers. In Beijing, this turnaround can be attributed to policy driven population control measures, which have yielded tangible outcomes through strict household registration management and the relocation of general manufacturing and wholesale markets, resulting in a sustained reduction of the permanent resident population. Tianjin, by contrast, continued its downward trajectory from the preceding phase, as sluggish economic recovery contributed to a persistent slight net population outflow. Meanwhile, five cities remained in the severe shrinkage category, representing 31.25% of the total. These include Baoding in central Hebei, Handan, Xinji, and Xingtai in southern Hebei, and Zhangjiakou in the northwest—all of which sustained the contractionary trends observed in the previous period, with the intensity of shrinkage further intensifying.

4.2 Population shrinkage in the YRD urban agglomeration

Based on population change rate data for the Yangtze River Delta (YRD) urban agglomeration spanning from 2010 to 2024, we conducted a statistical analysis of its population shrinkage status, as presented in Tables 3, 4 and illustrated in Figure 2.

Table 3. Cities experiencing population shrinkage in the YRD urban agglomeration.

Table 4. Number and proportion of cities with population shrinkage in the YRD urban agglomeration.

Between 2010 and 2015, population shrinkage in the YRD urban agglomeration occurred only in isolated cases within individual cities. Spatially, all shrinking cities were situated on the periphery of the region. Yancheng was the sole city experiencing mild shrinkage, accounting for 3.7% of the total. Located in the central coastal area of Jiangsu Province on the North Jiangsu Plain, Yancheng possessed extensive coastal land; however, its industrial development remained relatively underdeveloped during this period. Traditional agriculture constituted a substantial share of its economic structure, constraining its capacity to absorb population growth. Moreover, its remoteness from core metropolitan centers such as Shanghai and southern Jiangsu limited its ability to effectively benefit from industrial spillover effects, resulting in a modest net outflow of local population. Tongling was the only city experiencing severe population shrinkage, also accounting for 3.7%. Situated in south-central Anhui Province on the southern bank of the Yangtze River, Tongling is a typical resource-based industrial and mining city, with copper extraction and smelting as its dominant industries. During this period, as mineral resources gradually became depleted, traditional industries faced mounting pressures for transformation, leading to job losses and a pronounced population outflow.

Between 2015 and 2020, population shrinkage in the YRD urban agglomeration accelerated markedly, with its spatial scope expanding considerably. Geographically, this trend manifested as a belt like pattern extending westward along the Yangtze River and northward along the coastal corridor. The number of cities experiencing mild population shrinkage rose to 2—Chuzhou and Wuhu—accounting for 7.41% of the total. Both cities are situated in eastern Anhui Province, adjacent to Nanjing, and lie at the intersection of the Hefei and Nanjing metropolitan areas. Although they absorbed some industrial relocation during this period, they continued to face population siphoning pressure from core metropolitan centers such as Nanjing and Hefei, resulting in mild population decline. The number of cities experiencing severe population shrinkage surged to 6, representing 22.22% of the total, including Anqing, Chizhou, Ma’anshan, Taizhou, Xuancheng, and Yancheng. From a spatial perspective, these cities exhibit distinct belt like distributions. The first is the Wanjiang Urban Belt extending westward along the Yangtze River, encompassing Anqing, Chizhou, Ma’anshan, and Xuancheng—most of which are traditional industrial or resource based cities facing mounting pressures for industrial transformation and upgrading. The second consists of Taizhou and Yancheng on the northern coast of Jiangsu, located on the periphery of the YRD core area, where industrial development remains relatively lagging and persistent population outflow continues.

Between 2020 and 2024, population shrinkage in the YRD urban agglomeration exhibited new characteristics: the spatial extent of severe shrinkage contracted notably, whereas the scope of mild shrinkage expanded rapidly, giving rise to a pattern characterized by “broader coverage of mild shrinkage alongside residual pockets of severe contraction.” The number of cities experiencing mild shrinkage rose sharply to 6—Chizhou, Shanghai, Taizhou, Tongling, Xuancheng, and Yancheng—accounting for 22.2% of the total. The most noteworthy development during this phase is that Shanghai entered the mild-shrinkage category for the first time. As the core metropolitan center of the Yangtze River Delta, Shanghai was affected by a confluence of factors during this period, including the continued enforcement of population control policies, rising living costs, and industrial relocation, all of which contributed to a slight decline in its permanent resident population. Meanwhile, Chizhou, Taizhou, Xuancheng, and Yancheng—all of which had been classified as severely shrinking in the previous phase—transitioned to mild shrinkage, suggesting that the intensity of decline has abated, although population growth has not yet returned to positive territory. Tongling also shifted from severe to mild shrinkage, indicating that the transformation of this resource-based city has achieved partial success and that population outflow pressures have moderated. The number of cities in severe shrinkage plummeted to just one—Anqing—representing 3.7% of the total. Located in southwestern Anhui Province on the northern bank of the Yangtze River, Anqing is a major traditional industrial base and an important agricultural city. Compared with other shrinking cities, Anqing lies farther from the YRD core, possesses fewer locational advantages for attracting industrial relocation, and faces greater challenges in transforming its traditional industries, all of which have contributed to sustained and severe population outflow.

4.3 Population shrinkage in the mainland cities of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA)

In stark contrast to the BTH and YRD urban agglomerations, the mainland cities of the GBA experienced no population shrinkage in any of its constituent cities between 2010 and 2024. As shown in Table 5, all cities maintained positive growth over this period, reflecting a robust and sustained capacity for population aggregation. Benefiting from an integrated industrial layout, a well-developed transportation network, and strong economic vitality, the mainland cities of the GBA have formed a stable and cohesive population agglomeration system.

Table 5. Number and proportion of cities with population shrinkage in the mainland cities of the GBA.

5 Analysis of factors affecting population shrinkage in the three major urban agglomerations

5.1 Analysis of factors influencing population changes in the BTH urban agglomeration

5.1.1 Model fitting and data characteristics

This study constructs individual fixed-effects and two-way fixed-effects panel models to investigate the drivers of population change in the Beijing-Tianjin-Hebei urban agglomeration. The balanced panel dataset covers 15 prefecture-level cities over 15 years, comprising 225 valid observations. (Due to severe data gaps in Anyang, this city will be excluded from subsequent analyses.) From the fitting statistics, the within-group R² of the individual fixed-effects model is 0.121, which increases to 0.162 upon the inclusion of year fixed effects (two-way fixed effects). This indicates that incorporating time fixed effects moderately improves the model's explanatory power for intra-city population fluctuations. The overall F-test of both models yields Prob > F = 0.000, which is significant at the 1% level, confirming that the explanatory variables jointly possess significant explanatory power for the dependent variable, thereby validating the statistical soundness of the model specification.

To specify the model, we conduct the Hausman test. The result (χ² = 14.192, p < 0.05) rejects the null hypothesis in favor of the fixed effects model at the 5% level. Accordingly, we employ the fixed effects (FE) estimator in our main regressions.

5.1.2 Effects of core variables

In the individual fixed effects model (Table 6 Columns 1 and 2), only variable x2 is weakly significant at the 10% level with a negative coefficient of −0.092 (p = 0.164). All remaining explanatory variables (x1, x4, x5, x6, x7, x8, x10, x11, x13, x14) fail to pass the 10% significance threshold.The coefficient of x2 is negative, meaning that when controlling for time-invariant urban inherent characteristics, the year-on-year growth of x2 is associated with a slight decline in the population change index within the same city, though this correlation is weak and only marginally significant. For other non-significant variables.

Table 6. Results of the two-fixed effects regression for the three major urban agglomerations.

Standard errors in parentheses.

*p < 0.01, **p < 0.05, *p < 0.1.

x1 carries a negative coefficient (−0.05), implying that the natural population growth rate has a mild inhibitory effect on local population changes, yet the annual intra-city fluctuations of this indicator do not constitute a statistically stable driver. Economic and infrastructure variables (x4, x5, x6, x14) all present positive but insignificant coefficients. It indicates that the annual expansion of economic scale and municipal infrastructure does not effectively drive short-term population changes within individual cities; Transport-related indicators (x7, x8) have small, near-zero coefficients with high p-values. Railway construction and passenger flow volume exhibit no discernible linear association with population changes at the city-year level;(x10, x11, x13) are all negatively signed but insignificant, showing that the capacity of related public services to support population agglomeration fails to generate a stable effect when only annual city-level variations are considered.

This result is robust after eliminating nationwide annual macro shocks. Combined with the Hausman test result (x2 = 14.192, p = 0.048 < 0.05), we select the individual fixed effects model as the benchmark specification. The overall finding is clear: under the control of cities’ inherent constant features, almost none of the economic, infrastructural, and demographic indicators exert a statistically significant influence on intra-city population dynamics, with only x2 exhibiting a weak negative marginal effect. This reveals that the short-term annual population fluctuations of BTH cities are not primarily driven by the aforementioned socioeconomic factors, and inter-city static disparities—rather than yearly internal changes—dominate population differentiation across the region.

To mitigate multicollinearity, variables x3, x9, and x12—which exhibited high correlation—were excluded from the analysis. Subsequent VIF diagnostics for the BTH sample indicate a maximum VIF of 9.29, which is below the critical threshold of 10. This suggests only mild collinearity, with no material disturbance to the regression estimates.

5.2 Analysis of factors influencing population changes in the YRD urban agglomeration

5.2.1 Model fit and data characteristics

This study constructs individual fixed-effects and two-way fixed-effects panel models to identify the driving factors of population change across the Yangtze River Delta urban agglomeration. The unbalanced panel dataset covers 27 prefecture-level cities with a total of 405 valid observations, providing a sufficiently large sample for empirical estimation. The fitting statistics of the individual fixed-effects model (only city fixed effects controlled), where the within-group R² is 0.072. After the inclusion of year fixed effects (two-way fixed-effects model, Table 6 Columns 1 and 2), the goodness-of-fit increases marginally, although the overall explanatory power of the core variables for intra-city population fluctuations remains limited. The F-test statistic yields Prob > F = 0.000 for both specifications, which is significant at the 1% level, confirming the joint statistical significance of the full set of explanatory variables and validating the rationality of the model setting. A correlation test among indicators is not presented in this section. The regression results do not exhibit any apparent issues of multicollinearity, as the coefficients largely maintain consistent signs across the one-way and two-way fixed-effects models, with no abnormal distortion.

To specify the model, we conduct the Hausman test. The result (χ² = 15.928, p < 0.05) rejects the null hypothesis in favor of the fixed effects model at the 5% level. Accordingly, we employ the fixed effects (FE) estimator in our main regressions.

5.2.2 Effects of core variables

Table 6 Columns 3 and 4 further incorporates year fixed effects to eliminate nationwide macroeconomic shocks shared by all cities. The results show consistent sign trends for all variables across the two models, with only two indicators exhibiting statistical significance, while the remaining indicators fail to reach the 10% significance level.

In the individual fixed-effects model, the coefficient of x1 is −0.515 with p = 0.079, marginally significant at the 10% level. After controlling for year fixed effects, its negative coefficient expands to −0.661 but loses statistical significance. The negative sign implies that the natural population growth index cannot effectively boost intra-city annual population changes within the Yangtze Delta cities; natural demographic momentum does not serve as a core agglomeration driver when urban inherent conditions are controlled for.

This indicator exhibits a stable and statistically significant negative impact in both models. The coefficient stands at −0.002 (p < 0.01) under individual fixed effects and −0.00233 under two-way fixed effects, significant at the 1% level. It demonstrates that the yearly variation of x13 exerts a persistent restraining effect on urban population fluctuations, and this inhibitory influence is robust after eliminating common time shocks.

All remaining indicators (x2, x4, x5, x6, x7, x8, x10, x11, x14) are statistically insignificant at the 10% level in both regressions, with their economic implications interpreted as follows:

x2 carries a tiny negative coefficient close to zero, indicating its yearly fluctuations have a negligible linear correlation with population changes within individual cities; Economic indicators (x4, x5, x6) all yield negative but non-significant coefficients. Though regional economic development differs greatly across YRD cities, the annual growth of economic output does not generate stable short-term population agglomeration effects when only observing intra-city variation; Transportation variables (x7, x8) have small negative coefficients with high p-values. Railway infrastructure and passenger flow volume do not constitute a statistically reliable driving or crowding-out force on year-by-year population dynamics; Infrastructure indicator x10 presents a tiny positive coefficient yet lacks significance, meaning annual expansion of road construction does not translate into stable population attraction; public service (x11) and municipal investment (x14) yield small, mixed positive and negative coefficients that lack statistical significance, revealing that short-term annual investment and public service upgrades fail to exert a consistent effect on population adjustment within individual cities.

Overall, unlike the Beijing-Tianjin-Hebei sample with only weakly significant x2, the Yangtze Delta’s intra-city population fluctuations are mainly constrained by x13, while natural population growth only shows a marginal restraining effect without robust significance after controlling for macro annual factors. Most economic, transportation, and public service indicators cannot independently explain yearly population changes at the city level, implying that population differentiation across the YRD region is primarily determined by static inter-city gaps rather than annual internal factor adjustments.

Prior high-correlation variables x3, x9, and x12 have been eliminated in advance to address severe pairwise linear dependence among economic and land-use indicators. The post-filtering VIF test for the YRD sample confirms that the maximum VIF of all remaining explanatory variables is 9.34, with an average VIF of 5.28. All variance inflation factors are substantially lower than the critical threshold of 10, which means multicollinearity poses no substantial bias to the coefficient estimates of this sub-sample regression.

5.3 Analysis of factors influencing population changes in the mainland cities of the GBA urban agglomeration

5.3.1 Model fit and data characteristics

This paper constructs one-way individual fixed-effects and two-way fixed-effects panel models to explore the driving factors of population change in the mainland cities of the GBA. The balanced panel dataset comprises 9 prefecture-level cities over 15 consecutive years, yielding a total of 135 valid balanced observations. The estimation results of the individual fixed-effects model (only city fixed effects controlled), whose within-group (R²) is 0.239. After introducing year dummy variables to construct the two-way fixed-effects model (Table 6 Columns 5 and 6), the goodness-of-fit increases to 0.447. The F-test of both regression models obtains Prob > F = 0.000 at the 1% significance level, which confirms that the explanatory variables jointly possess explanatory power for population change, and the model specification is statistically valid. This section does not report correlation matrix results, and no apparent coefficient distortion due to multicollinearity is observed in the regression outputs.

To specify the model, we conduct the Hausman test. The result (χ² = 22.273, p < 0.05) rejects the null hypothesis in favor of the fixed effects model at the 5% level. Accordingly, we employ the fixed effects (FE) estimator in our main regressions.

5.3.2 Effects of core variables

Table 6 Columns 5 and 6 further adds year fixed effects to eliminate common macro shocks across all cities. All explanatory variables maintain consistent sign directions in the two models, yet only two indicators show statistical significance in the individual fixed-effects model, and all variables become insignificant after controlling for year fixed effects.

The coefficient of x5 is −0.001 with (p = 0.021), significant at the 5% level. It indicates that within a single city, the annual increase in x5 is correlated with a decline in the population change index. However, this inhibitory effect dissipates after accounting for common cross-year macroeconomic shocks, suggesting that the negative correlation is driven by periodic macro-trends rather than stable internal mechanisms.

The coefficient of x8 is negative (p = 0.049), also significant at the 5% level under individual fixed effects. It shows that passenger flow expansion restrains annual population adjustment when only city fixed effects are controlled for. Once common annual macro factors are accounted for, this correlation loses its statistical robustness.

All other variables (x1, x2, x4, x6, x7, x10, x11, x13, and x14) fail to reach the 10% significance level in both one-way and two-way fixed-effect regressions, with economic interpretations as follows:

Demographic indicator x1 has negative coefficients in both tables without significance, which implies natural population growth does not generate a sustained population agglomeration effect for annual intra-city fluctuations. Economic indicators (x2, x4, x6) carry mixed positive and tiny negative coefficients with high p-values. Even with large economic gaps among Bay Area cities, year-on-year economic expansion does not generate stable short-term population attraction;

Transportation variable x7 shows an insignificant negative coefficient, demonstrating that high-speed rail construction exerts no reliable linear influence on yearly population dynamics. Infrastructure and public service variables (x10, x11, x13, x14) all lack statistical significance, meaning short-term annual investment and service upgrades do not effectively adjust a city's yearly population scale.

5.4 Analysis of regression results using integrated panel data for the three major urban clusters

5.4.1 Model fit and data characteristics

This study pools panel data covering three major urban agglomerations and constructs two-way fixed-effects models to estimate the overall influencing effects, with interaction terms introduced to conduct cross-group coefficient heterogeneity tests. The full sample includes 51 prefecture-level cities and 765 valid balanced observations. Table 7 reports the benchmark two-way fixed-effects regression for the pooled sample, with a within-group R² of 0.062. After incorporating grouping interaction terms to capture regional disparities, the goodness-of-fit rises slightly to 0.085 (Table 8). The F-statistics of both regressions yield Prob > F = 0.000, which rejects the null hypothesis that all coefficients are jointly equal to zero at the 1% significance level and verifies the rationality of model specification. No correlation matrix results are reported in this section, and no severe coefficient distortion due to multicollinearity is observed in the regression outputs. The categorical variable x15 is adopted to categorize samples by urban agglomeration, where (x15 = 1), representing the mainland cities of the GBA, serves as the reference group, (x15 = 2) denotes the Yangtze River Delta, and (x15 = 3) represents the Beijing-Tianjin-Hebei region. Interaction terms constructed by x15 and each explanatory variable capture the differences in marginal effects relative to the benchmark group. The regression results reveal that only one interaction term shows statistical significance, suggesting limited cross-regional heterogeneity for most driving factors.

Table 7. Results of the two-fixed effects regression for the three major urban agglomerations.

Standard errors in parentheses.

*p < 0.01, **p< 0.05, *p< 0.1.

Table 8. Heterogeneity results for the three major urban agglomerations.

Standard errors in parentheses.

*p < 0.01, **p < 0.05, *p < 0.1.

After removing highly collinear indicators x3, x9, and x12, we conducted a VIF test for the pooled sample of the three urban agglomerations. The maximum VIF of all retained covariates is 6.19, and the mean VIF equals 2.77, both well below the critical threshold of 10, confirming that multicollinearity does not significantly bias the pooled regression estimates.

5.4.2 Unified impact effects across the full sample

Table 7 presents the benchmark two-way fixed-effects regression results for all three urban agglomerations. Only two explanatory variables reach conventional levels of statistical significance, while all other indicators exhibit statistically insignificant correlations with population change at the 10% level. First, the coefficient of x1 is −0.388 and significant at the 1% level. This finding suggests that, on average across all sample cities, higher natural population growth exerts a restraining effect on annual intra-city population fluctuations, after controlling for city and year fixed effects. Second, x13 carries a significantly negative coefficient of −0.00152 at the 1% level, demonstrating a persistent inhibitory impact of this indicator on population dynamics within cities. For the remaining variables (x2, x4, x5, x6, x7, x8, x10, x11, x14), their coefficients carry mixed positive and negative signs yet lack statistical significance. Economic development, high-speed rail infrastructure, passenger flow, road construction, medical resources, and municipal investment fail to generate universally stable promoting or restraining effects on population change across the three major urban agglomerations, with no consistent cross-regional driving mechanism being identified.

5.4.3 Effects of regional heterogeneity (based on intergroup coefficient tests)

Table 8 reports the estimation results of the heterogeneity model embedded with group-variable interaction terms, with the mainland cities of the GBA (x15 = 1) as the reference group. Each interaction term reflects the gap in marginal effects between the corresponding urban agglomeration and the benchmark region. Among all interaction terms corresponding to explanatory variables, only (3.x15#c.x13) yields a coefficient of −0.00205, significant at the 5% level. All other interaction terms associated with (x1, x2, x4, x5, x6, x7, x8, x10, x11) and x14 fail to reach the 10% significance cutoff.This evidence indicates that the inhibitory effect of x13 on population change is significantly stronger in the BTH region than in the GBA. In contrast, the marginal impacts of natural demographic factors, economic output, transportation facilities, and public service provisions do not differ statistically across the three urban agglomerations. Overall, the underlying logic of population variation remains largely homogeneous across the sample regions, and regional differentiation only exists in the constraint channel of variable x13.

5.5 Endogeneity tests

Endogeneity bias, stemming from omitted variables and reverse causality, may distort the benchmark regression estimates. To address this concern, this study employs the first-order lag of the core variable x1 (L.x1) as the instrumental variable and conducts instrumental variable regression for robustness verification. The lagged explanatory variable satisfies the exogeneity requirement: it is correlated with the contemporaneous endogenous regressor but exerts no direct influence on the dependent variable after controlling for full sets of covariates and time fixed effects. Table 9 reports the instrumental variable estimation results. The regression covers 51 cities with a total of 714 valid observations, and the within-group R² is merely 0.035. All explanatory variables, including the lagged core variable L.x1, fail to pass the 10% significance threshold. Among all year dummy variables, only the dummy variable for 2020 exhibits marginal significance at the 10% level, with a negative coefficient of −2.686, whereas the rest of the year fixed effects display insignificant coefficients. Comparing the instrumental variable results with the baseline two-way fixed-effects regression, the sign of x1 remains negative after mitigating potential endogeneity issues, which is consistent with the benchmark results. Nevertheless, the coefficient of L.x1 loses statistical significance, implying that the restraining impact of natural population growth on population fluctuations is not robust once endogeneity concerns are addressed. Other economic, transportation, and public service indicators still exhibit no stable linear correlation with the explained variable, which is consistent with the previous pooled regression outcomes. In summary, after alleviating endogeneity bias through lagged instrumental variable estimation, the overall explanatory power of the model is relatively low, and almost all of the proposed driving factors lack statistical significance. Only the 2020 time dummy presents weak negative constraints on population change, which may be attributable to short-term external shocks in that particular year. Although the sign direction of the core explanatory variable remains unchanged, its significance disappears, suggesting that the significant inhibitory effect of x1 detected in the baseline model is partially attributable to potential endogeneity.

Table 9. Results of endogeneity tests.

Standard errors in parentheses.

*p < 0.01, **p < 0.05, *p < 0.1.

6 Discussion

This paper systematically investigates population change patterns and their driving mechanisms across the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and the mainland cities of the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) from 2010 to 2024. Combined with segmented sub-sample regressions, pooled full-sample estimation, cross-group heterogeneity tests and endogeneity robustness checks, the empirical results both complement and refine existing conclusions, while also revealing distinct regional mechanisms that have not been identified in prior literature.

Beyond these methodological contributions, the empirical results reveal three distinct trajectories of urban shrinkage that reflect fundamentally different development models across the three urban agglomerations. The BTH region exhibits a top-down shrinkage path, driven primarily by proactive policy interventions—Beijing's population control measures and the systematic relocation of non-capital functions have actively reshaped population distribution, pushing even the core metropolises into mild shrinkage. In contrast, the YRD follows a bottom-up shrinkage path, where market forces—industrial restructuring, rising living costs in core cities, and the locational disadvantages of peripheral areas—drive population redistribution. Notably, shrinkage intensity in the YRD shows signs of easing, as most cities previously classified as severely shrinking have transitioned to mild shrinkage. The mainland cities of the GBA represent a resilient growth path, where deep regional integration, industrial complementarity, and sustained economic vitality have maintained positive population growth across all cities. These contrasting patterns suggest that urban shrinkage within agglomerations is not a uniform phenomenon, but rather a context-dependent outcome shaped by the interplay of policy regimes, market forces, and regional integration levels. The coexistence of these three distinct paths within China's most developed regions underscores the necessity for place-based rather than one-size-fits-all policy responses.

6.1 Reassessment of regional population change drivers relative to prior studies

Previous relevant studies have generally identified multiple stable and statistically significant promoting or restraining factors for urban population fluctuations, covering demographic, economic, transportation, infrastructure, and ecological dimensions. However, sub-sample two-way fixed-effect regressions in this paper deliver divergent new findings, thereby establishing a targeted dialogue with the existing literature.

For the BTH urban agglomeration, only x2 shows marginal negative significance at the 10% level under individual fixed effects, while all other economic, transportation and public service indicators lack statistical robustness. This result revises the mainstream view that passenger flow or high-speed rail constitutes a core restrictive force in northern urban clusters (Deng et al., 2019; Ma et al., 2024). The weak explanatory power of conventional socioeconomic variables indicates that short-term annual population adjustments within BTH cities are not primarily driven by yearly changes in economic output or transportation accessibility. The so-called "siphoning" and "outflow" effects identified in earlier aggregate analyses largely reflect static inter-city gaps rather than dynamic annual factor variations, which complements the differentiated logic of policy and market forces shaping population patterns in core versus peripheral cities proposed by Jin and Sui (2021).

In the YRD sample, only x1 is weakly significant at the 10% level in the one-way fixed-effects model, and x13 maintains stable significantly negative effects across both one-way and two-way specifications. All remaining economic, transport, and public service indicators are insignificant. This finding differs from Ma and Yoon (2025), who documented multifaceted bidirectional impacts of industrial restructuring and demographic growth. The empirical evidence here demonstrates that yearly fluctuations in industrial output, high-speed rail construction and ecological land scale cannot reliably explain intra-city population volatility within the Yangtze River Delta. The restraining role of x13 becomes the sole consistent short-term constraint factor, implying that previous conclusions regarding multiple drivers may be sensitive to model specification and sample composition.

For the mainland cities of the GBA, two indicators (x5, x8) display significant negative correlations only under individual fixed effects, and both effects vanish after controlling for common year fixed effects. No variables retain statistical significance in the two-way fixed model. This revises Yang et al. (2024), who argued high-speed rail and natural population growth generate sustained population agglomeration effects in bay areas. The transience of the significant coefficients for x5 and x8 suggests that periodic macro-trends, rather than stable internal mechanisms, drive temporary population restraints in the mainland cities of the GBA. Even though the region maintains overall population vitality, annual adjustments in economic and transportation factors fail to form persistent forces shaping city-level population dynamics.

6.2 Regional heterogeneity revisited based on pooled regression and cross-group coefficient tests

Prior multi-region comparative research presumed widespread cross-group disparities in the marginal effects of core driving variables (Ma et al., 2023). Nevertheless, the pooled full-sample two-way fixed-effects regression and the heterogeneity interaction tests in this paper provide more nuanced evidence.

The benchmark pooled regression of all three urban agglomerations identifies merely two robust variables: x1 and x13 are significantly negative at the 1% level, whereas all other economic, transport, and public service indicators show no consistent universal impact across the entire sample. This means there is no unified set of driving factors applicable across all three major urban clusters, challenging the notion of a generalizable multi-factor framework proposed in most single-region studies.

The cross-group coefficient heterogeneity test adopts the mainland cities of the GBA as the reference group, with interaction terms capturing effect gaps for the BTH and YRD. Among all interaction items corresponding to explanatory variables, only (x13) reaches the 5% significance level, indicating that the inhibitory effect of x13 is markedly stronger in the BTH than in the mainland cities of the GBA. All other demographic, economic, and transportation indicators yield insignificant cross-group interaction coefficients. Such a result greatly weakens the overstated regional differentiation in existing literature. While the three urban agglomerations differ in policy background, industrial structure and spatial layout, most factor channels shaping annual population changes operate in a statistically homogeneous manner; regional divergence is concentrated solely in the constraining pathway of x13. This supplements the scale differentiation theory of Ma et al. (2023) by demonstrating that systematic inter-cluster heterogeneity is limited even at the same prefecture-city administrative scale.

6.3 Robustness implications from endogeneity instrumental variable estimation

To mitigate potential estimation bias from reverse causality and omitted variables, this paper employs the one-period lag of x1 (L.x1) as an instrumental variable for endogeneity testing. The IV regression yields a within-group (R²) of merely 0.035, with nearly all explanatory variables (including the lagged core variable (x1)) losing statistical significance. Only the year dummy of 2020 is marginally negative at the 10% level.

Comparing instrumental variable outcomes with the baseline pooled regression: the coefficient sign of x1 remains negative after correcting for endogeneity, yet its statistical significance disappears. This implies the significant restraining effect of natural population growth detected in the original full-sample model is partially attributable to endogeneity. All other socioeconomic indicators stay insignificant under both specifications, further confirming that annual population fluctuations across cities cannot be robustly explained by yearly variations in conventional driving factors once endogenous risks are eliminated. This outcome provides a critical supplementary interpretation for prior empirical studies that omitted endogeneity checks, highlighting that apparent significant correlations in static fixed-effect models may lack causal validity.

7 Conclusions and policy implications

7.1 Main conclusions

Using panel data from 2010 to 2024, this study identifies the spatiotemporal characteristics of population shrinkage across three major urban agglomerations—the Beijing-Tianjin-Hebei (BTH) region, the Yangtze River Delta (YRD), and the mainland cities of the Guangdong–Hong Kong–Macao Greater Bay Area (GBA)—and employs a two-way fixed-effects panel regression model to analyze its determinants. The core conclusions are as follows.

Population shrinkage across the three urban agglomerations exhibits significant regional heterogeneity. In the BTH region, shrinkage evolved from sporadic distribution across individual cities during 2010–2015 to a widespread trend during 2015–2020, ultimately forming a comprehensive coverage pattern by 2020–2024, with the core cities of Beijing and Tianjin also entering the category of mild shrinkage. The YRD is characterized by mild shrinkage spreading across a broad area, while severe shrinkage remains concentrated in isolated pockets. From 2020 to 2024, the core city of Shanghai experienced mild shrinkage for the first time, whereas the intensity of shrinkage in most cities previously classified as severely shrinking has notably eased. Throughout the entire study period, none of the mainland cities of the GBA experienced population shrinkage; all maintained positive growth, demonstrating a strong and sustained capacity for population aggregation.

Regarding influencing factors, population changes in the BTH region were significantly and negatively affected only by passenger traffic volume. Under the siphoning effect of the core cities, there was a marked outflow of population from peripheral areas, while the supportive roles of economic and infrastructure factors were not fully realized. In the YRD, population changes were driven by multiple factors: the natural population growth rate, per capita GDP, and road area exerted significant positive influences, whereas the GDP of the tertiary sector, the number of high-speed rail stations, and passenger traffic volume imposed significant negative constraints, reflecting issues such as industrial agglomeration crowding out and transportation-induced outflow. In the mainland cities of the GBA, the natural population growth rate and the number of high-speed rail stations served as the core positive drivers, while passenger traffic exerted a significant negative influence. The high degree of regional integration means that indicators such as GDP and urbanization rates do not exhibit significant effects.

Regression analysis of the full sample confirmed the positive effects of natural population growth, public services, and infrastructure investment on population agglomeration, while also verifying the negative impacts of passenger traffic volume and tertiary industry GDP. Intergroup coefficient tests further revealed significant regional differences among the three urban agglomerations regarding the effects of natural population growth, economic development, and transportation infrastructure. Additionally, endogeneity tests validated the reliability of these conclusions. Overall, the differences in spatiotemporal characteristics and impact mechanisms of population shrinkage across the three major urban agglomerations stem from variations in regional development positioning, industrial structure, policy regulation, and the interplay of siphoning and radiation effects among cities.

7.2 Policy implications

Based on the research findings, the following policy implications are proposed.

First, differentiated population governance strategies should be implemented. In response to the pronounced siphoning effect observed in the BTH region, it is essential to strengthen the radiative influence of the core metropolitan centers. This can be achieved through the relocation of non-capital functions and the cultivation of secondary central cities, thereby fostering a multi-centered, networked urban structure that alleviates the pressure of unidirectional population outflow. In light of the multi-factor-driven characteristics of the YRD region, industrial structures and transportation networks should be optimized to guide rational population mobility and to prevent localized shrinkage from being exacerbated by excessive industrial agglomeration or transport-induced spillover effects. For the mainland cities of the GBA, continued efforts should be made to leverage their existing advantages in regional integration so as to sustain and consolidate their momentum of population aggregation.

Second, public service and infrastructure support should be strengthened. The full-sample regression results confirm that investment in public services and infrastructure exerts a significantly positive effect on population agglomeration. Accordingly, fiscal investment in foundational sectors—including education, healthcare, and road networks—should be increased to enhance cities’ comprehensive carrying capacity and livability, with particular attention to addressing the deficiencies in public service provision in shrinking cities, so as to improve their population attractiveness. At the same time, the negative effects of factors such as passenger traffic volume must be rationally evaluated, and the comprehensive transportation system should be optimized to strike a balance between transportation accessibility and population stability.

Third, regional coordination and endogenous growth should be promoted. Given the significant positive effect of natural population growth rates, it is necessary to improve the policy framework for childbirth support, alleviate the financial burdens on families in childbearing and education, and foster long-term balanced population development. Each urban agglomeration should build upon its own functional positioning to drive industrial transformation and upgrading, as well as coordinated regional development. Specifically, the BTH region should accelerate the transition from traditional to new drivers of economic growth to unlock its full economic potential; the YRD should guide the orderly transfer and sharing of industries; and the mainland cities of the GBA should continue to function as a growth pole through high-level openness. Ultimately, these efforts are expected to achieve spatial equilibrium and sustainable development across population, economy, and the resource-environment nexus.

8 Research limitations and future prospects

This study examines population shrinkage in the three major urban agglomerations—the BTH region, the Yangtze River Delta, and the mainland cities of the GBA. Although relevant conclusions have been drawn, several limitations remain due to constraints related to data and methodology. First, the mainland cities of the GBA were exclusively analyzed owing to the unavailability of data for Hong Kong and Macao, which limits a comprehensive reflection of the region's demographic characteristics. Second, all research data were sourced from official statistical yearbooks and bulletins, without support from micro-level survey data. Consequently, it was not possible to analyze the mechanisms underlying individual-level population mobility decisions, resulting in an insufficient explanation of micro-level drivers. Third, population shrinkage was identified solely using the annual average change rate of the permanent resident population, without integrating indicators from the economic, industrial, or public service dimensions. As a result, it was difficult to comprehensively and objectively characterize the true state of urban shrinkage, potentially overlooking cities with latent shrinkage. Fourth, using prefecture-level cities as the sole unit of analysis, without extending to district and county levels, prevented identification of variations in population shrinkage across hierarchical levels within urban agglomerations.

In addition, this study employs two-way fixed-effects models to identify the local driving effects of annual socioeconomic factors on intra-city population growth rates. Still, it does not conduct spatial autocorrelation tests (e.g., Moran’s I) or spatial econometric estimations such as spatial Durbin models. Consequently, cross-city spatial spillover and siphoning effects are not quantitatively decomposed in this paper. Future research could construct geographic weight matrices, calculate global and local Moran’s I statistics, and adopt spatial panel models to systematically quantify the interactive population linkages among neighboring cities, thereby further enriching the research framework for urban population shrinkage in major urban agglomerations.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Yongwan Chun, The University of Texas at Dallas, United States

Reviewed by: Pingjun Sun, Southwest University, China

Xuan Hou, Chinese Academy of Sciences (CAS), China