The Reconstruction of China’s Population Mobility Pattern Under Digital Technology Evolution: A Pathway to Urban Sustainability

Abstract

Population mobility is increasingly crucial for regional development. However, current studies often neglect the impact of rapid digitalization. This study adopts a three-stage analytical framework derived from the Techno-Economic Paradigm across its incubation, penetration, and maturity phases to examine how digital technology evolution has reshaped China’s population mobility patterns. Through ERGM and social network analysis, we found the following: (1) During the incubation period (1980s–2000), digital technology enhanced economies of scale, leading to a siphoning effect of the population from inland to coastal areas. (2) In the penetration phase (2000–2017), digital technology had a dual effect. Automation weakened coastal agglomeration by replacing labor, while the digital industry created new inland clusters of employment, ultimately reshaping population mobility into a multi-center structure. (3) In the maturity phase (2018–present), the concentration of skilled workers in technology hubs and the dispersal of displaced labor to less digitally advanced areas formed a multi-centered and networked population mobility pattern, thereby enhancing the sustainability and spatial balance of the urban system through functional specialization and the matching of skill profiles to city roles.

1. Introduction

The sustainable development of urban systems constitutes a central challenge globally, with population mobility playing a pivotal role in shaping economic vitality, social equity, and environmental outcomes. In China, large-scale population mobility in the early stage of reform and opening up not only solved the problem of labor shortage in coastal areas, but also triggered a demographic dividend [1,2], which has become a key mechanism for maintaining China’s long-term economic growth. However, after decades of rapid development, data released by the National Bureau of Statistics in 2022 revealed that China’s natural population growth rate became negative (−0.6‰) for the first time since 1960 (−4.57‰). Meanwhile, the proportion of the population aged 65 and above reached 14.2% in 2021, which indicates that society has already entered an aging phase. Therefore, the traditional development model relying on population expansion is no longer feasible.

Studies show that regions that can attract highly skilled mobile populations can accumulate human capital and form a new demographic dividend, while regions with insufficient mobility may face a decline in growth momentum [3,4]. Therefore, the strategic governance of population mobility has risen to a key competitive advantage in the urban sustainability framework.

Research on population mobility has become a hot topic in recent years. The determinants of this phenomenon have been investigated in research areas such as job opportunities and housing market conditions [5,6,7], and analysis has been carried out on how mobile populations address challenges in cultural adaptation, social welfare accessibility, and community network formation [8,9,10,11]. In addition, researchers have studied mobile population attributes such as mobility motivation [12,13,14,15], and the effectiveness of policies that aimed to attract new residents has been evaluated [16,17]. Finally, others have analyzed individual mobility decisions using discrete choice models such as Logit models [18,19,20].

Existing studies on population mobility have made significant progress, but certain limitations remain. China has undergone a remarkable evolution of digital technology since 1980 [21,22,23], when digital technologies started to see their first applications in some industrial scenarios. From 2000 to 2017, e-commerce, mobile payment, and the sharing economy flourished, and since 2018, digital technology has been integrated into various industries of the real economy. However, this tendency might be the key variable in reshaping China’s population mobility pattern. Research has indicated that technological progress in the digital realm differs fundamentally from previous advancements. Historical industrial revolutions advanced gradually, fostering incremental growth and creating widespread employment. In contrast, the new industrial revolution driven by digital technology is rapidly disrupting existing occupations. While it fosters new industries and enhances labor productivity, it also brings negative impacts like workforce displacement and industrial hollowing-out [24,25,26,27]. These transformations inevitably influence regional wage levels and employment opportunities, thereby altering population mobility decisions. In fact, existing studies have demonstrated that digital technologies can influence inter-firm innovation networks, transportation networks, and social networks [28,29,30]. However, this factor has not been incorporated as a core variable in traditional models analyzing population mobility networks. Meanwhile, research has shown an excessive focus on individual decision-making behaviors. Therefore, it predominantly relies on traditional econometric regression models and fails to combine emerging methodologies such as Exponential Random Graph Modeling (ERGM) and spatial econometrics.

On this basis, this study aims to analyze the mechanisms through which digital technology evolution has reshaped China’s population mobility patterns across different phases, thereby tracing how this restructuring fosters the sustainability of the urban system. To achieve this, we employ cutting-edge methods. The remainder of this paper is organized as follows: Section 2 presents the research hypotheses, Section 3 details the research design, Section 4 reports the empirical results, and Section 5 provides the conclusion and policy implications.

2. Research Hypotheses

The Techno-Economic Paradigm, which was proposed by Dosi in 1982, provides an excellent theoretical perspective for this study. This theory categorizes technological evolution into three distinct phases, incubation, penetration, and maturity [31,32], each of which drives profound transformations in both economic and social structures [33].

Based on this paradigm, this study divides the development trajectory of China’s digital technology into three evolutionary stages.

During the first phase (1980s to early 21st century), the main feature of digital technology was its niche industrial applications, such as finance and telecommunications, where computers only carried out basic data functions.

Next, we analyze the impact of digital technology on population mobility patterns using the following neoclassical production function:

$$\mathrm{Y}=\mathrm{A}\mathrm{f}\left(\mathrm{K},\mathrm{L}\right)$$

In the above equation, Y represents total output, A denotes total factor productivity (TFP), and K and L stand for capital input and labor input, respectively.

In this stage, digital technology mainly functions as Hicks-neutral technological progress [34,35], increasing A while preserving the original capital–labor ratio (K/L), thus satisfying the following:

$${\displaystyle \frac{\partial \left(K/L\right)}{\partial A}}=0$$

Next, we examine the digital technology-driven productivity premium. According to marginal productivity theory, the wage level of labor is given by the following:

$$\mathrm{w}=\mathrm{P}\mathrm{r}\times {\mathrm{M}\mathrm{P}}_{\mathrm{L}}=\mathrm{P}\mathrm{r}\times \mathrm{A}\times \left(\partial \mathrm{f}/\partial \mathrm{L}\right)$$

In the equation, Pr represents the price level of output, MP_L denotes the marginal product of labor, and ∂f/∂L is the partial derivative of the production function with respect to labor.

It is important to note that digital technology will directly enhance TFP (A), thereby increasing the marginal product of labor. The increase in wage levels (w) will generate a siphoning effect, as coastal areas with technological advantages continue to attract labor from digitally underdeveloped inland areas.

From the beginning of 2000 to 2017, China’s digital technology development entered an accelerated growth track and permeated all fields of manufacturing and services [36].

This phase exhibits a fundamental transformation of the production function, evolving from the Hicks-neutral form to the capital-augmenting form. This structural shift can be expressed as follows:

$$\mathrm{Y}=\mathrm{f}\left(\mathrm{A}·\mathrm{K},\mathrm{L}\right)$$

In this formulation, A · K represents capital embedded with digital technologies, such as industrial robots and CNC machine tools. Compared with the previous stage’s Hicks-neutral technological progress that increased MP_K and MP_L proportionally, capital-augmenting technological progress will widen the ratio of MP_K to MP_L, resulting in a rise in the relative return on capital compared to labor [37,38].

Next, we examine the crowding-out mechanism in the labor market. A firm’s cost-minimizing decision satisfies the following:

$${\displaystyle \frac{{\mathrm{M}\mathrm{P}}_{\mathrm{K}}}{{\mathrm{M}\mathrm{P}}_{\mathrm{L}}}}={\displaystyle \frac{r}{w}}$$

In the equation, r represents the cost of capital and w denotes wages. As analyzed above, the advancement of digital technologies will cause the MP_K to rise. From the above formula, it can be seen that when r remains unchanged, workers’ wages (w) will decrease, leading to capital (such as CNC equipment) replacing industrial labor. Meanwhile, due to new employment opportunities such as IT positions in the field of digital research and development, regions at the forefront of digital technology development will drive new population inflows. These twin forces will not only dilute the agglomeration effect in regions previously dominated by manufacturing but also fuel the rise of new demographic hubs in digitally advanced cities, prompting a structural shift toward a multi-center population mobility network.

During the mature stage (2018–present), digital technology has penetrated all sectors of the economy (Perc et al., 2019) [39].

In this context, the production function further evolves into the following:

$$\mathrm{Y}=\mathrm{f}\left(\mathrm{A}\mathrm{I}·\mathrm{K},\mathrm{L}\right)$$

Here, AI · K represents intelligent capital, while L represents labor utilizing AI-assisted tools.

The capital-augmenting effect becomes more pronounced as newer generations of digital technology like AI demonstrate stronger empowerment capabilities.

On one hand, smart manufacturing will accelerate the replacement of labor, which would lead to a continuous decline in the population-attracting capacity of traditional industries. Thus, people will flow to regions where digital technology is relatively lagging. On the other hand, the vigorous development of the digital economy will create countless new job opportunities, driving the flow of high-quality labor forces toward digital industry agglomeration zones. This dual mechanism of substitution and creation will become more pronounced than in the previous stage. The consequent weakening of traditional industries’ agglomeration effect will displace mid- and low-skilled laborers toward less digitally advanced cities. Meanwhile, digitally ascendant cities will attract a growing share of high-skilled talent. Consequently, the population mobility network will evolve from its previously multi-center structure toward a more networked pattern, with individuals of different skill levels gravitating to functionally distinct cities.

This spatial reorganization, driven by functional specialization and skill-to-role matching, redistributes development opportunities from a few coastal megacities to a broader network of cities. By doing so, it alleviates the environmental and social burdens of over-concentration. Simultaneously, it activates economic vitality by creating distinct development pathways for different cities, thereby enhancing the long-term sustainability of the urban system.

Based on the preceding analysis, this study proposes the following hypotheses:

H1. 

During the incubation phase, digital technology generated a siphoning effect, drawing population from inland regions to coastal areas.

H2. 

In the penetration phase, it reshaped population mobility into a multi-center structure.

H3. 

Throughout the maturation phase, digital technology further transformed population flow into a multi-centered and networked pattern.

3. Research Design

3.1. Data Sources

This study constructed population mobility networks based on two authoritative datasets.

The first dataset was the China Mobility Dynamic Survey (CMDS) initiated by the National Health Commission of China in 2009 and continued until 2017. This nationwide survey mainly studies the mobile populations aged 15–59 residing in destination locations for over one month. Through random sampling methodology, it gathers annual data from approximately 200,000 mobile households.

Subsequently, we constructed population mobility networks by aggregating individual mobility decisions to form them at an intercity scale, as shown in Figure 1.

The second dataset comprises population mobility data for 367 Chinese cities (Amap, June 2018–December 2023). Municipal socioeconomic variables came from the China City Statistical Yearbook, supplemented by the CEIC database.

The application of these two datasets in this study is as follows. Since the CMDS was discontinued after 2017, while Amap data became available starting in 2018 with no overlap between the two sources, the population mobility networks for the incubation phase (pre-2000) and penetration phase (2000–2017) were constructed using CMDS data, whereas Amap data were employed for the maturity phase (post-2018).

To ensure the comparability of networks constructed from different data sources, we conducted a robustness test. In the model for the maturity phase, we incorporated the 2017 population mobility network (PopMob2017) built from CMDS data as a network covariate to test path dependency. A significantly and positive coefficient for this covariate indicates that despite differences in data collection methods, the macro-level network patterns and core node rankings revealed by both datasets show strong consistency, suggesting that comparative studies using different data sources are feasible. The results are presented and discussed in Section 4.3.

3.2. Social Network Analysis

Social network analysis is a core tool for this study. It is employed to delineate the structure of population mobility networks and to evaluate the importance of nodes (cities) within them.

On one hand, we employed weighted in-degree centrality, where the higher the value, the stronger the city’s ability to attract population inflows within the network [40,41]. To facilitate cross-network comparisons, weighted in-degree is typically normalized to a [0, 1] scale using the following formula:

$$C_{in}\left(i\right)={\displaystyle \frac{{\sum }_{j=1}^n{w}_{ji}}{max\left({\sum }_{k=1}^n{w}_{ki}\right)}}$$

where w_ji represents the weight of the edge from nodes j to i. The denominator is the maximum weighted in-degree value among all nodes in the network.

This study also adopted weighted out-degree centrality to measure the extent of population outflow from a city node in the network. Its value equals the sum of weights of all edges directed from this node to others. Normalization was applied using the following formula:

$$C_{out}\left(i\right)={\displaystyle \frac{{\sum }_{j=1}^n{w}_{ij}}{max\left({\sum }_{k=1}^n{w}_{ik}\right)}}$$

where w_ij represents the weight of the edge from nodes i to j, and the denominator is the maximum weighted out-degree value among all nodes in the network.

3.3. ERGM Analysis

Exponential Random Graph Modeling (ERGM) is a method for analyzing the drivers of network structure formation, which has broad applicability. For example, one study employed this approach to examine the impacts of market and institutional factors on intercity investment networks [42]. Other research investigated the driving mechanisms of innovation network evolution, using China’s photovoltaic industry and new energy vehicle industry as case studies, respectively [43,44].

The general form of the model is expressed as

$$\mathrm{P}\left(\mathrm{Y}=\mathrm{y}|\mathrm{X}\right)={\displaystyle \frac{1}{\mathrm{ᴋ}}}\mathrm{e}\mathrm{x}\mathrm{p}\{{\sum }_{\mathrm{k}}{\theta }_{\mathrm{k}}{\mathrm{z}}_{\mathrm{k}}\left(\mathrm{y}\right)+{\sum }_{\mathrm{p}}{\beta }_{\mathrm{p}}{\mathrm{h}}_{\mathrm{p}}\left(\mathrm{y},\mathrm{X}\right)+{\sum }_{\mathrm{q}}{\gamma }_{\mathrm{q}}{\mathrm{g}}_{\mathrm{q}}\left(\mathrm{y},\mathrm{W}\right)\}$$

where Y represents the random network; the parenthetical terms denote independent variables; θ, β, and γ are parameters corresponding to the variables; and ᴋ is the normalization constant.

Unlike regression analysis, ERGM explains network formation mechanisms through three categories of variables (see

Table 1. Main variables of ERGM model and their definitions.

Parameter	Structure	Explanation
Endogenous Structural Variables
Edges		Represents the baseline connection likelihood. (+): denser than random; (−): sparser than random.
Mutual		Measures reciprocity in directed networks (A→B and B→A). (+): higher mutual connection preference than random. (−): stronger unidirectional tendency than random.
Node Attribute Variables
Receiver Effect		Measures differential ability of nodes to attract connections. (+): Significant attraction/clustering tendency in specific nodes. (−): Connection avoidance pattern in particular nodes.
Network Covariates
Exogenous Network Effect		Measures cross-network influence on tie formation. (+): External networks systematically strengthen specific connections. (−): External factors suppress particular ties.

): Endogenous structural variables test whether observed network configurations stem from intrinsic structural mechanisms. Node attribute variables assess how individual characteristics influence connections. In particular, a positive Receiver Effect shows that certain nodes attract significantly more incoming ties. Network covariates evaluate external influences on tie formation.

3.4. Indicator Selection

This study incorporates three types of explanatory variables including endogenous structural factors, node attributes, and network covariates (see

Table 2. Indicator selection and explanation of the ERGM model.

Variable Type	Variable Name	Definition
Dependent Variable	Population mobility network	Directed weighted network matrix constructed from intercity population flow data.
Endogenous Structural Variables	Edges	Baseline connection density of the network.
	Mutual	Reciprocal flow relationship.
Node Attribute Variables (Receiver Effect)	Manu	Manufacturing development level of a city.
	Serv	Service sector development level of a city.
	Digital	Digital technology development level of a city.
	Digital × Manu	Product term of Digital and Manu (multiplicative interaction).
	Digital × Serv	Product term of Digital and Serv (multiplicative interaction).
Network Covariates	GeoProx	Geographical proximity effect (inverse Euclidean distance between cities).
	CulProx	Cultural proximity effect (1 if cities share a dialect region, 0 otherwise).
	PopMob2017	The 2017 CMDS-based network, used to test path dependence and validate comparability with the 2023 Amap-based network.

for details).

First, the dependent variable is the population mobility network. We analyze the 2000, 2017, and 2023 networks for the incubation (pre-2000), penetration (2000–2017), and maturation (post 2023) periods, respectively.

Second, the core explanatory variables are Manu and Serv, measured by the revenue of designated scale enterprises. Both are expected to positively affect population inflows, as industrial expansion creates jobs and enhances the attractiveness of cities. Digital is measured using telecommunications revenue as the foundational metric. Given that digital technologies evolve across different developmental stages, a single indicator can no longer adequately capture their multidimensional characteristics. Therefore, for the penetration phase, we augmented this foundation with the number of patent grants and employed the entropy weight method (EWM) to construct a normalized comprehensive index. For the maturity phase, we further incorporated the number of national high-tech enterprises and also applied the EWM to calculate a normalized comprehensive index. This indicator is also expected to have a positive effect due to its capacity to generate employment and attract population. The interaction terms Digital × Manu and Digital × Serv are expected to have negative effects, indicating that digital technology may displace traditional jobs and weaken the pull effect of industrial expansion.

Third, geographical and cultural proximity are included to capture their influence on mobility networks. Meanwhile, PopMob2017 is aimed to test path dependence and validate the comparability between the CMDS and Amap data sources, as described in Section 3.1. In addition, Edges and Mutual have been defined in the preceding section and thus is not reiterated here.

4. Empirical Results

4.1. Analysis of the Digital Technology Incubation Phase

During the incubation period of digital technology before 2000, the development of digital technology in China was relatively slow. During this period, China mainly carried out preliminary digital transformation of traditional industries through computer technology [45,46]. Due to the limited prevalence of the Internet, digital applications exhibited localized characteristics and had not yet achieved deep integration across industrial ecosystems. The applications included enterprise inventory management software, early office automation (OA) systems, and computerized accounting systems.

In this phase, the population mobility network shows a strong siphoning effect (see

Table 3. Centrality rankings of Chinese cities in the national population mobility network (2000).

Ranking	Weighted In-Degree (Normalized)		Weighted Out-Degree (Normalized)
	City	Value	City	Value
1	Shanghai	1	Chongqing	1
2	Beijing	0.901	Fuyang	0.525
3	Tianjin	0.744	Suihua	0.504
4	Dalian	0.307	Qiqihar	0.440
5	Wuxi	0.224	Xinyang	0.440

).

Based on weighted in-degree centrality, four coastal cities rank in the top five, including Shanghai (1.000), Beijing (0.901), Tianjin (0.744), and Dalian (0.307). Meanwhile, weighted out-degree centrality is highest in inland cities, including Chongqing (1.000), Fuyang (0.525), Suihua (0.504), Qiqihar (0.440), and Xinyang (0.440).

Figure 2 further illustrates these spatial patterns. Coastal areas have emerged as major population centers, with Shanghai, Beijing, and Tianjin distinctly marked in red. Adjacent regions show a gradient transition from orange to yellow, while the vast inland areas remain uniformly deep green, indicating weaker population attraction.

The ERGM estimation results for the incubation stage of digital technology are shown in

Table 4. ERGM estimation results for the digital technology incubation phase.

	Model 1	Model 2
Endogenous Structural Variables
Edges	−3.5119 *** (0.0242)	−4.1131 *** (0.0317)
Mutual	0.9207 *** (0.0869)	−0.1215 (0.1111)
Node Attribute Variables
Manu	4.4868 *** (0.4381)	4.5107 *** (0.4653)
Serv	5.9833 *** (0.5072)	6.7518 *** (0.5601)
Network Covariates
GeoProx	—	264.6704 *** (10.0459)
CulProx	—	0.9096 *** (0.0590)
Model Fit Statistics
AIC	22,654.3669	20,679.5700
BIC	22,690.7751	20,734.1822
Log-likelihood	−11,323.1835	−10,333.7850
Number of Nodes	258	258

Note: *** p < 0.001; standard errors are shown in parentheses; “—“ indicates the variable was not included in the model.

. Model 1 analyzes the effects of endogenous network structures combined with Manu and Serv. Model 2 introduces additional network covariates including GeoProx and CulProx. Comparative analysis shows that Model 2 has superior performance, as evidenced by its higher log-likelihood value (−10,333.79) and lower information standard value (AIC: 20679.57; BIC: 20734.18).

Model 2 reveals a non-random mobility network, evidenced by a significantly negative edge coefficient (β = −4.1131, p < 0.001). The non-significant Mutual coefficient (β = −0.1215) confirms unidirectional siphoning flows. Second, the significant positive coefficients of Manu (β = 4.5107, p < 0.001) and Serv (β = 6.7518, p < 0.001) point to industrial scale expansion as a driver of population agglomeration. Finally, the positive effects of GeoProx and CulProx underscore the roles of geography and culture in mobility decisions.

The analysis reveals that during the digital technology incubation phase, population mobility exhibited the coastal siphoning effect, thereby validating Hypothesis H1.

4.2. Analysis of the Digital Technology Penetration Phase

From the beginning of 2000 to 2017, the application of digital technologies in China evolved from single-point uses to comprehensive systemic implementations. According to statistics from the World Intellectual Property Organization (WIPO), the number of PCT patent applications in China surged to third place in the world with 21,516 in 2013, and further rose to 43,168 by 2016, which had reached comparable levels with the US and Japan (see Figure 3).

During this period, coastal areas showed a weakened siphoning effect, while inland areas exhibited emerging multi-nodal development characteristics (see

Table 5. Centrality rankings of Chinese cities in the national population mobility network (2017).

Ranking	Weighted In-Degree (Normalized)		Weighted Out-Degree (Normalized)
	City	Value	City	Value
1	Beijing	1	Fuyang	1
2	Shanghai	0.926	Zhoukou	0.994
3	Tianjin	0.737	Shangrao	0.780
4	Zhengzhou	0.392	Suihua	0.769
5	Nanjing	0.376	Bijie	0.751

).

Analysis of weighted in-degree centrality shows that although traditional first-tier cities such as Beijing (1.000) and Shanghai (0.926) continue to dominate, several inland regional centers including Zhengzhou (0.392) and Nanjing (0.376) have advanced into the top five rankings. Meanwhile, the weighted out-degree centrality analysis reveals that the top five cities are all located in central and western China, with Fuyang (1.000) and Zhoukou (0.994) ranking highest.

Figure 4 further demonstrates the restructuring of population mobility patterns. While traditional core cities like Beijing and Shanghai remain deep red, multiple new orange clusters have emerged nationwide. These phenomena indicate that coastal cities retain agglomeration advantages, while emerging digital hubs in inland regions develop new attraction poles.

The ERGM results (

Table 6. ERGM estimation results for the digital technology penetration phase.

	Model 3	Model 4	Model 5	Model 6
Endogenous Structural Variables
Edges	−2.2564 *** (0.0147)	−2.2891 *** (0.0150)	−2.5036 *** (0.0168)	−3.1332 *** (0.0208)
Mutual	0.6547 *** (0.0336)	0.6547 *** (0.0336)	0.6613 *** (0.0346)	0.0402 (0.0404)
Node Attribute Variables
Manu	9.1190 *** (0.3693)	0.4438 (0.5053)	2.6028 *** (0.5820)	0.0351 (0.6169)
Serv	0.6450 *** (0.0292)	−0.1013 *** (0.0264)	1.0137 *** (0.0441)	1.2013 *** (0.0491)
Digital	—	7.6891 *** (0.2366)	9.5261 *** (0.2318)	10.1640 *** (0.2386)
Digital × Manu	—	—	−37.0079 *** (1.8736)	−33.7121 *** (1.9485)
Digital × Serv	—	—	−0.8212 *** (0.0742)	−1.1107 *** (0.0790)
Network Covariates
GeoProx	—	—	—	450.8189 *** (8.9248)
CulProx	—	—	—	0.4410 *** (0.0377)
Model Fit Statistics
AIC	63,940.7131	63,081.8155	61,630.3544	56,220.3811
BIC	63,977.9471	63,128.3579	61,695.5138	56,304.1574
Log-likelihood	−31,966.3566	−31,535.9077	−30,808.1772	−28,101.1905
Number of Nodes	286	286	286	286

Note: *** p < 0.001; standard errors are shown in parentheses; “—“ indicates the variable was not included in the model.

) show four progressively specified models (Models 3 to 6). Model 6 demonstrates the best fit, with the highest log-likelihood (−28,101.19) and lowest AIC (56,220.38) and BIC (56,304.16) values. Its key findings are shown below:

First, the continuing insignificance of the Mutual (β = 0.0402, p > 0.01) confirms the unidirectional flow pattern from inland to coastal regions. Second, Manu demonstrates a non-significant effect (β = 0.0351, p > 0.05), while Serv shows a significant positive effect (β = 1.2013, p < 0.001). These results confirm that traditional service sector expansion remains a driver of population inflows, whereas the agglomeration effect of manufacturing expansion on employment has weakened. However, the interaction terms between digital technology and Manu and Serv reveal substantial negative coefficients (Digital × Manu: β = −33.7121, p < 0.001; Digital × Serv: β = −1.1107, p < 0.001), indicating that digital penetration has eroded the employment-generating capacity of traditional industries and further restrained population inflows. Third, Digital demonstrates a positive effect (β = 10.1640, p < 0.001), indicating that cities prioritizing digital technology development have generated substantial employment opportunities and thus enhanced their attractiveness.

The analysis has validated H2. The capital-augmenting technological progress driven by digital technologies exhibits labor substitution effects. As equipment such as robots becomes embedded in production processes, the marginal product of capital (MP_K) rises relative to the marginal product of labor (MP_L), leading firms to substitute capital for labor and thereby weakening the siphoning effect of industrial expansion in coastal regions on population. Meanwhile, by creating new employment opportunities, the digital industry itself is transforming inland tech hubs like Chengdu into new attraction centers. This dual mechanism of substitution and creation has facilitated the formation of a multi-center population mobility network.

4.3. Analysis of the Digital Technology Maturity Phase

The period since 2018 has been defined by the rapid integration of AI, blockchain, cloud computing, and big data (often grouped as ABCD). This evolution is fundamentally reshaping industry, pushing operations beyond simple automation toward truly intelligent systems.

According to the Digital China Development Report (2024) (see Figure 5), a staggering 84.1% of major industrial enterprises now use digital R&D tools, up from 73.5% just four years ago. What is perhaps even more significant for manufacturing is that the numerical control rate for critical processes jumped from 54.7% to 66.4% in the same period.

At the same time, China’s population mobility pattern has undergone significant changes (see

Table 7. Centrality rankings of Chinese cities in the national population mobility network (2023).

Ranking	Weighted In-Degree (Normalized)		Weighted Out-Degree (Normalized)
	City	Value	City	Value
1	Guangzhou	1	Guangzhou	1
2	Dongguan	0.877	Dongguan	0.869
3	Shenzhen	0.805	Foshan	0.743
4	Foshan	0.730	Shenzhen	0.710
5	Suzhou	0.615	Suzhou	0.602
…	…	…	…	…
9	Chengdu	0.394	Wuxi	0.403

).

From the perspective of weighted in-degree centrality, Chengdu (0.394) entered the top ten, breaking the traditional gap between the east and west. Simultaneously, the weighted out-degree centrality analysis shows that economically vibrant cities such as Guangzhou (1.000) and Dongguan (0.869) rank at the top. This dual prominence in both inflow and outflow demonstrates the emergence of bidirectional mobility patterns, which manifests the characteristics of networked development.

Figure 6 further provides some interesting changes. In the eastern coastal region, agglomerations like the Yangtze River Delta show a distinct color transition from red to orange. Two transformative trends have emerged. First, the attractiveness in inland regions has significantly increased, which formed multiple orange nodes. Second, except for the northern and some western ethnic regions, most cities appear in shades of light green to yellow, with a notable reduction in color differences between them. This evidence indicates that China’s population mobility is transitioning toward a polycentric and networked structure.

The ERGM analysis for the digital technology maturity phase (

Table 8. ERGM estimation results for the digital technology maturity phase.

	Model 7	Model 8	Model 9	Model 10	Model 11
Endogenous Structural Variables
Edges	−1.9946 *** (0.0192)	−1.9832 *** (0.0195)	−2.1690 *** (0.0196)	−4.8670 *** (0.0365)	−4.9173 *** (0.0400)
Mutual	3.7137 *** (0.0305)	3.7116 *** (0.0312)	3.7239 *** (0.0294)	2.5978 *** (0.0327)	2.5656 *** (0.0337)
Node Attribute Variables
Manu	1.5039 *** (0.0313)	0.6761 *** (0.0475)	0.9085 *** (0.0484)	0.4654 *** (0.0557)	0.5428 *** (0.0607)
Serv	0.0340 (0.0429)	−0.3472 *** (0.0272)	−1.5681 *** (0.2242)	2.4606 *** (0.2652)	1.3420 *** (0.2875)
Digital	—	7.4302 *** (0.3775)	15.4051 *** (0.5174)	14.6269 *** (0.5737)	14.9486 *** (0.6349)
DT × Manu	—	—	−4.0444 *** (0.1214)	−2.9955 *** (0.0934)	−3.1275 *** (0.1127)
DT × Serv	—	—	2.1732 *** (0.2798)	−2.5475 *** (0.2659)	−1.4284 *** (0.2938)
Network Covariates
GeoProx	—	—	—	3948.1368 *** (43.7004)	3954.3563 *** (46.2312)
CulProx	—	—	—	0.5127 *** (0.0680)	0.5021 *** (0.0720)
PopMob2017	—	—	—	—	0.8272 *** (0.0374)
Model Fit Statistics
AIC	78,440.9942	77,892.0808	76,390.9922	50,799.3721	46,907.8737
BIC	78,478.2839	77,938.6930	76,456.2493	50,883.2741	47,000.5337
Log-likelihood	−39,216.4971	−38,941.0404	−38,188.4961	−25,390.6860	−23,443.9368
Number of Nodes	288	288	288	288	280

Note: *** p < 0.001; standard errors are shown in parentheses; “—“ indicates the variable was not included in the model.

) examines a sequence of four progressively specified models (Models 7 to 10). The model comparison results demonstrate that the comprehensive Model 10 provides the best statistical fit among all specifications, as evidenced by its maximized log-likelihood value (−25,390.69) and minimized information criteria scores (AIC = 50,799.37; BIC = 50,883.27). In addition, Model 11 is a robustness test that introduces the 2017 population mobility network as a network covariate to test path dependency. The sample size was reduced from 288 to 280 cities due to the selection of cities common to both the CMDS and Amap datasets.

Model 10 shows a significant positive mutual effect (Mutual: β = 2.5978, p < 0.001), signaling a shift from unidirectional concentration to reciprocal interactions. While Manu (β = 0.4654, p < 0.001) and Serv (β = 2.4606, p < 0.001) maintain significant positive effects, their interactions with Digital are significantly negative (Digital × Manu: β = −2.9955, p < 0.001; Digital × Serv: β = −2.5475, p < 0.001). This points to a strong labor substitution effect by digital technology in traditional sectors, reducing their pull on population. Meanwhile, Digital has an unprecedented positive effect (β = 14.6269, p < 0.001), establishing it as the primary driver of population agglomeration. As discussed in the theoretical section, the capital-augmenting effect of digital technologies has intensified. Mature technologies like AI and big data are displacing traditional labor more profoundly, which redirects ordinary labor flows toward inland and lower-tier cities with slower digital adoption. Concurrently, the digital sector has become a powerful demographic magnet, drawing in high-skilled talent toward innovation hubs. The significantly positive coefficient of Mutual further validates this emerging dual-direction mobility pattern. Ultimately, driven by these twin forces, China’s population mobility network has restructured into a multi-center, networked structure. Hypothesis H3 is thereby substantiated.

Finally, as demonstrated by Model 11 (robustness test), the significantly positive coefficient of PopMob2017 (p < 0.001) confirms that the network structure from the penetration phase can predict mobility patterns in the maturity phase. This result provides robust evidence for the consistency of macro-level network patterns and core node rankings revealed by both datasets, conclusively supporting the feasibility of a diachronic comparative analysis.

5. Conclusions and Policy Implications

This study applies social network analysis and ERGM to examine how digital technology evolution reshaped population mobility patterns in China.

The findings are as follows: First, in the incubation stage of digital technology, superior returns on capital and labor triggered population movement from inland to coastal areas. Second, during the penetration phase, digital technology improved the marginal output efficiency of capital, triggering labor outflow in industrial clusters while simultaneously creating numerous job opportunities. This dual dynamic promoted the formation of a polycentric mobility structure. Third, in the maturation phase, labor displacement intensified in traditional industries, while more new industries were spurred. As a result, the relocation of medium-skilled workers to inland cities and the concurrent concentration of high-end talent in coastal cities collectively shaped a multi-nodal networked mobility pattern. Consequently, the maturity phase was thus instrumental in building a more sustainable and spatially balanced urban system for China. In this system, the synergy between centers of talent attraction and regions that absorb general labor alleviated the burdens of monolithic agglomeration while fostering cross-regional economic vitality.

To adapt to these shifting patterns, this study proposes the following measures: First, given regional disparities in digital technology [47,48], a monitoring mechanism should be established to track digital technology levels across regions, thereby providing reliable support for implementing differentiated policies. Second, in areas with severe population decline, net population inflow could be incorporated into the KPIs of local officials. Third, considering the divergence in population flows, with high-end talent accumulating in first-tier cities and ordinary workers moving toward lower-tier cities [39,49,50], targeted policies are needed to balance spatial development. First-tier cities like Beijing and Shanghai should provide support for professionals in AI, big data, and cloud computing. At the same time, vocational training systems must be strengthened to enhance the adaptability of ordinary workers to the digital economy. Such a dual approach is essential for achieving a sustainable and coordinated urban system that enables cities with distinct functions to attract and retain residents of all skill levels.