Introduction

Amidst the new wave of technological revolution and industrial change, exploring innovative means to improve production efficiency has emerged as a critical issue that needs to be addressed urgently in contemporary society. Along with the Internet, big data, cloud computing, and the Internet of Things (IoT), artificial intelligence (AI) has continually expanded its applications across various fields, with robotics emerging as a novel form of intelligent equipment. Reports from the Ministry of Industry and Information Technology suggest that robotic technology now spans 60 industrial categories and 168 subsectors of the national economy, profoundly influencing both production processes and daily lifeFootnote 1. As the demographic dividend fades, escalating labor costs are compelling an increasing number of firms to adopt sophisticated production technologies, notably including robotics. In 2022, China experienced a 5% increase in robotic installations, reaching 290,258 units, constituting 52% of the global aggregateFootnote 2. While this surge in robotic deployment has markedly enhanced production efficiency and expanded operational scale, it has also raised concerns about potential job risks (Frey and Osborne, 2017; Javed, 2023). Currently, the Chinese labor market faces constraints arising from an aging population and diminishing fertility rates. This demographic shift signals the gradual decline of the traditional demographic dividend, posing risks not only to employment stability but also affecting social efficiency and equity. Enhancing the adaptability and stability of employment can facilitate market-oriented resource allocation and enhance coordinated regional development, in which maintaining employment stability of the rural migrant workforce directly correlates with the stable increase in the income levels of the poverty-stricken population, the high-quality employment of the rural labor force, and the development of labor market integration. The rural floating population represents a significant demographic segment, with China experiencing a population increase from 221 million in 2010 to 376 million in 2020, marking an average annual growth rate of 8.3% over this periodFootnote 3. In addition, the scope of rural labor mobility is expansive, with data from China’s 2020 population census displaying the country’s inter-municipal mobile population at 376 million. Of which, the inter-provincial mobile population accounted for 125 million, while the intra-provincial mobile population reached 251 millionFootnote 4. While this substantial migrant workforce has been pivotal to China’s economic and social progress, it faces considerable pressure to achieve stable employment. Therefore, exploring the impact of robot applications on the migrant population holds practical significance.

Unlike previous advancements in technology, the effects of robot applications on the labor market remain uncertain. On one hand, robots eliminate certain repetitive and labor-intensive jobs, increasing the risk of unemployment for those in low-skill positions (Acemoglu and Restrepo, 2020). On the other hand, robot applications can enhance production efficiency and facilitate the expansion of production scales, yielding a positive impact on the labor market (Graetz and Michaels, 2018; Huang et al., 2022; Wang et al., 2024a). The rural labor workforce, predominantly engaged in low-tech and monotonous tasks, is more significantly affected by such disruptions. This particular demographic may face unemployment, be compelled to transition into alternative sectors, or relocate in pursuit of new employment opportunities. Previous studies have largely focused on local labor markets, overlooking the impact on labor’s spatial mobility. Academic research on labor migration encompasses various dimensions including technological progress (Yuan and Pan, 2023), policy interventions (Chen et al., 2019b), migration costs (Lin et al., 2022), transportation infrastructure (Han and Kung, 2015), and unique urban characteristics (Luong et al., 2023). While studies such as Brougham and Haar (2020), which utilized data from the United States, Australia, and New Zealand, have noted that the adoption of robotics tends to steer the workforce towards job transitions, these explorations mainly cover developed countries, often falling short in comprehensively analyzing the effects of robotics on labor mobility. The movement of labor across regions represents not only an individual’s pursuit of maximizing utility but also a reallocation of labor resources. Therefore, investigating labor mobility across regions is crucial for achieving a rational distribution of labor, fostering high-quality development, and enhancing living standards.

This paper explores the rapidly expanding domain of robot applications, focusing particularly on their implications for employment stability, the migration of rural laborers, and the underlying mechanisms driving this process. This study also examines the effects of robotic applications on income distribution and production efficiency. Utilizing the ‘Bartik IV’ approach, a comprehensive index system is used to measure urban robot stock density and analyze the secondary sector workforce using data from the China Migrant Dynamic Survey (CMDS) for 2014–2018.

Our findings suggest that a 1% increase in urban robot application density is correlated with a 0.249% increase in the likelihood of rural laborers contemplating relocation, as validated by several robustness tests. Urban robot applications were found to reduce the migration rate of urban laborers and increase the probability of rural labor returning to the agricultural sector. Probing the mechanisms underlying the initial regression results, our results showed a notable passive crowding effect attributable to robotic applications. The impact of robotic applications on the willingness of laborers to transfer is further evidenced by an increase in working hours. The effect of robot applications on accelerating rural labor transfer is particularly pronounced among lower-skilled workers, individuals over 44, those in low-skill jobs, highly mobile demographic groups, and those in economically developed areas. Further analyses suggest that robot application contributes to an increase in regional average wages (a 1% increase in urban robot density corresponded to a 0.086% increase in the average wage level of rural laborers). In terms of production efficiency, the utilization of robots was observed to enhance the productivity of urban labor.

The primary contributions of this article are as follows: First, the current literature has mainly focused on local labor markets, while the impact of robot applications on resource allocation has largely been overlooked. Given the significant number of rural migrants in China and their more pronounced response to robotics technology, this study focuses on the rural migrant population. By quantitatively analyzing the impact of robot applications on the rural labor market, particularly on the crowding-out effect, this approach provides a novel perspective on understanding the broader implications of robot applications on labor market dynamics. Second, this study diverges from the majority of existing studies that typically assess the overall impact of robot applications on the labor market, based primarily on the perspectives of labor skills, gender, and hukou (household registration) status. In this study, a micro-level individual perspective is adopted for deeper insights. The crowding-out effect can then be differentiated into active and passive types, based on variables such as employment status, insurance participation, and migration intentions. This nuanced approach is employed to explore the mechanisms through which robot applications influence rural labor migration. The results can provide substantive guidance for government policy in order to refine the employment mechanisms for the migrant population, encourage a systematic and orderly migration of rural labor, and ensure stability in the rural labor employment market. Third, given the significant heterogeneity in labor skill demands arising from robot applications, this study adopted and refined the simplified task framework proposed by Acemoglu and Restrepo (2022). But instead of solely categorizing tasks into labor-intensive and robot-intensive segments, our approach allows the consideration of the influence of robotic technological advancements on task variety, integrating the workforce skill levels into the task model. Our study systematically analyzes the impact of robot applications on labor demand, exploring expansion effects and substitution effects. In addition, inspired by the labor migration framework proposed by Chen et al. (2022), this study comprehensively examines the impact of robot applications on labor mobility, providing a theoretical foundation for analyzing the broader implications of robot applications on the labor market.

The rest of this paper is arranged as follows: The literature review section systematically explores existing research on the rural labor market, labor mobility, and the application of robots, establishing a foundation for further analysis. In the theoretical analysis section, labor mobility and robot application are integrated into a unified theoretical framework, and relevant hypotheses are formulated to guide the empirical investigation. The research design section outlines the current state of rural labor mobility and the developmental trends within the robotics industry, defining essential variables for the study and conducting a descriptive analysis of these variables. In the empirical research section, the impact of robotics applications on rural labor mobility is discussed, and the results are tested for robustness. The mechanisms through which robot applications influence labor mobility are analyzed from various angles, including transfer willingness, skill level, mobility, regional economic development, and income. In the further analysis section, the investigation is expanded to explore the effects of robot applications on specific vulnerable groups and the allocation of factors. The paper concludes with the conclusions and policy recommendations section, wherein the main findings are summarized and policy suggestions are proposed regarding the future direction of the robotics industry and the enhancement of employment promotion mechanisms for rural labor.

Literature review

China’s class stratification diverges significantly from that of the West, primarily due to the intricate interplay between the dual economic structure and the dual social system. This has resulted in a socially stratified framework marked by multiple divisions, as manifested by the segregation stemming from the urban-rural divide. Within this context, research has focused mainly on three main areas: First, the inequality in the factor exchange between urban and rural areas has received considerable research focus. For example, Meng and Zhao (2018) highlighted the disadvantaged position of rural areas in factor allocation, directly impacting the efficiency of allocating resources like capital, labor, and technology. Second, research has highlighted the imbalanced allocation of public resources between urban and rural areas. For instance, Cao et al. (2024), through their analysis of Nanjing, China, found the effects of municipal transportation and urban enterprise density on the development of the urban-rural interface. Lastly, the impediment in urban-rural circulation has become a major research focus. In one study, Liu et al. (2013) analyzed the urban-rural development patterns in China from 1996 to 2009 and pinpointed policies, institutions, and urbanization as critical drivers of changes in the spatial-temporal urban-rural landscape.

Given the growing emphasis on the challenges facing rural development in China, scholars have explored various issues, including rural labor force employment (Zhang et al., 2021), rural land rights (Bu and Liao, 2022), rural poverty reduction (Zhang et al., 2023), the urban-rural income disparity (Wang et al., 2024b), and resource misallocation (Chari et al., 2021). Amidst the evolution of emerging technologies, the substantial size and relatively low educational level of China’s rural workforce underscore the impediments to employment security. Examining the effects of robotic technology on the mobility of rural labor would be crucial to improving the mechanisms that promote rural labor employment and ensuring the stability of rural labor employment.

Labor mobility transfer factors

Labor mobility is essential to China’s economic growth and productivity enhancement, offering a potential solution to rural labor poverty. Imbalances in regional resource allocation have necessitated labor migration, which can broaden employment opportunities for local workers and mitigate the downward wage pressure on the non-mobile population exerted by the mobile population (Hong and McLaren, 2015). Research on labor migration has explored various dimensions such as technological advances, policy interventions, migration costs, and urban characteristics (Su et al., 2018; Lewis and Peri, 2015). For example, from a technological advancement perspective, Yuan and Pan (2023) used data from publicly listed companies and found that digital technology optimized labor allocation and expanded the unconventional workforce. Chen et al. (2019a) suggested that the gig economy had a significant influence on labor employment choices and their spatial distribution. In terms of policy interventions, An et al. (2024) utilized China’s 2014 household registration reform policy and concluded that it removed barriers to labor mobility between cities without negatively impacting local workers’ wages. In another study, urban identity was found to influence labor mobility (Chen et al., 2019b). Regarding migration costs, Combes et al. (2015) identified a positive relationship between the proportion of urban migrants and the wages of local residents. Gee et al. (2017) concluded that social networks provided employment opportunities for newcomers. In terms of urban characteristics, Chen et al. (2022) found that air pollution reduced immigration and increased labor emigration. Han and Kung (2015) argued that investments in urban infrastructure and services influenced labor mobility decisions. Other researchers have focused on the impact of natural disasters on labor migration (Li, 2024; Luong et al., 2023; Naoi et al., 2020).

However, the existing literature has largely failed to fully consider the impact of emerging technologies, such as robot applications, on rural labor migration behavior. This research gap has prevented a comprehensive assessment of the impact of robot applications on the labor market.

The impact of robot applications on the labor market

Academic discussions regarding the impact of robot applications on labor force employment have presented varying perspectives. On the one hand, scholars have suggested that robots can reduce labor demand through substitution effects, particularly in routine job sectors (Goddard et al., 2021; Frey and Osborne, 2017). Acemoglu and Restrepo (2020), in a study examining robot penetration metrics at the commuting zone level in the U.S., found that robot applications lowered both regional employment and wage levels. Analyzing data from a Chinese household survey, Giuntella et al. (2022) concluded that robot applications decreased household labor participation rates, reduced hourly wages, and lengthened working hours. Brambilla et al. (2023) suggested that robot applications displaced young and semi-skilled workers, resulting in lower employment rates. On the other hand, others have posited that robot applications stimulate labor demand through expansion effects, primarily manifested in increased production scales and the creation of new jobs (Lin, 2009; Mokyr et al., 2015). Autor (2015) observed that automation in the industrial sector had both destructive and creative effects on employment, with the complementarity between automation and human labor leading to increased demand for workers. Hjort and Poulsen (2019), using labor data from South Africa, discovered a notable correlation between robot applications and increased labor demand by enterprises.

In addition, some scholars have argued that robot applications do not significantly alter regional employment rates. Graetz and Michaels (2018) concluded that robot applications enhance economic productivity, thereby raising labor wages without markedly affecting the overall employment rate. Dauth et al. (2018), from a labor transfer perspective, noted that while robot applications did not affect overall employment figures, they facilitated the transfer of labor from manufacturing to service sectors.

Research on the effects of robot applications across different skill levels indicates that robot adoption may lead to job losses and decreased wages for low-skilled workers while simultaneously increasing demand and boosting wages for high-skilled labor (Dustmann et al., 2017; Huber and Stephens, 2014). Blanas et al. (2019) using data from developed countries, found that robot adoption reduced the demand for low and medium-skilled labor, along with female workers, leading to a shift towards more skill-intensive employment structures. De Vries et al. (2020) concluded that the increased use of robots had a negative correlation with employment in non-routine jobs, coupled with a decline in routine employment, although these effects were not significant in emerging countries and transition economies.

In this paper, the impact of robot applications on rural labor markets in China is explored in order to provide a better understanding of how robots affect labor dynamics.

Theoretical analysis and research hypothesis

This study utilizes the task model developed by Acemoglu and Restrepo (2022), which divides production tasks into segments specifically designated for labor and robots. The model is adapted to accommodate a diverse labor force, integrating varying skill levels into the framework and explicitly accounting for the effects of robotic technological advancements on task allocation. In addition, this paper investigates the job creation effects of robot applications and provides a detailed analysis of their impact on labor demand. Following the labor mobility framework outlined by Chen et al. (2022), the influence of robot applications on labor migration is further examined, and theoretical hypotheses are formulated. The analysis is conducted within the framework of a perfectly competitive market, ensuring that market clearance conditions are met and that the behavior of representative enterprises is thoroughly examined.

The impact of robot applications on labor demand

Assuming that the representative firm’s production function for tasks takes the Cobb-Douglas form, the equation can be expressed as follows:

$${Y}_{dt}=exp \left({\int }_{N-1}^{N}{ln}({y}_{{dt}}({s})){ds}\right)=exp \left({\int }_{N-1}^{N}{ln}\left({A}_{{dt}}^{s}({x}_{{dt}}({s}))^{1-\alpha }{K}_{{dt}}({s})^{\alpha }{ds}\right.\right)$$
(1)

where \({Y}_{dt}\) is the output level of representative enterprise d at time t, achieved through a series of consecutive production tasks \([N-1,N]\); \({A}_{{dt}}^{s}\) is the productivity coefficient; \({K}_{{dt}}\left(s\right)\) is capital stock; \({x}_{{dt}}(s)\) is the input of production task s. Note that \(0 \,<\, \alpha \,<\, 1\). For simplicity, each task is standardized to 1, \(s\in [\mathrm{0,1}]\). Consequently, Eq. (1) can be expressed as follows:

$${Y}_{dt}={\exp} \left({\int }_{0}^{1}{ln}\left({A}_{{dt}}^{s}({x}_{{dt}}({s}))^{1-\alpha }{K}_{{dt}}({s})^{\alpha }{ds}\right.\right)$$
(2)

where \(x(s{)}_{{it}}\) represents labor input required during the production process, with different sectors employing various skill types. Upon integrating robots into the model, they are capable of substituting for certain low-skilled labor tasks. \({{R}_{{dt}}^{K},W}_{dt}^{H},{W}_{{dt}}^{L}\) denote the prices of capital stock, high-skilled labor, and low-skilled labor, respectively. \({R}_{{dt}}^{M}\) reflects the cost of robot usage, covering both ownership and leasing expenses such as installation, depreciation, and maintenance. \({R}_{{dt}}^{K}\) represents the price of capital stock. The productivity of high-skilled labor, low-skilled labor, and robots are represented by \({r}_{{it}}^{H}\left(s\right),{r}_{{it}}^{L}\left(s\right),{r}_{{it}}^{M}(s)\), respectively. Under the assumption of fixed wages and considering the business production process, the model adheres to the relationship \({W}_{{dt}}^{H}/{r}_{{dt}}^{H}\, >\, {W}_{{dt}}^{L}/{r}_{{dt}}^{L}\, >\, {R}_{{dt}}^{M}/{r}_{{dt}}^{M}\). Following the principle of cost minimization, Eq. (2) can be formulated as follows:

$${y}_{{dt}}(s)=\left\{\begin{array}{c}{r}_{{dt}}^{M}(s){(m}_{{dt}}(s{))}^{1-\alpha }{K}_{{dt}}{(s)}^{\alpha },s\in [0,{I}^{* }]\\ {r}_{{dt}}^{L}(s)({l}_{{dt}}(s{))}^{1-\alpha }{K}_{{dt}}{(s)}^{\alpha },s\in ({I}^{* },\theta ]\\ {{r}_{{dt}}^{H}(s)(h}_{{dt}}{(s))}^{1-\alpha }{K}_{{dt}}{(s)}^{\alpha },s\in (\theta ,1]\end{array}\right.$$
(3)

where \({h}_{dt}\left(s\right),{l}_{dt}\left(s\right),{m}_{dt}\left(s\right),{k}_{{dt}}(s)\) denote the inputs of high-skilled labor, low-skilled labor, robots, and capital for task s, respectively. I* represents the technological frontier of automation, defining the scope of tasks that can be automated using technologies such as robots. Tasks exceeding I* cannot be automated. θ refers to the skill gap between high-skilled and low-skilled labor. Tasks with values greater than θ require high-skilled labor, as they are beyond the substitution capability of low-skilled labor. Let \({I}^{* }=I-(\theta -I){e}^{-{r}_{{dt}}^{M}(s)}\), with I signifying the lower threshold for automation in task s. If \({I}^{* }=\theta\), this means that robots can entirely replace low-skilled labor. Defining the demand for robots, low-skilled labor, and high-skilled labor for task s, and considering the properties of the Cobb–Douglas production function, it can be represented as follows:

$${k}_{{dt}}(s)=\alpha \frac{{Y}_{{dt}}}{{R}_{{dt}}^{K}}$$
(4)
$${m}_{{dt}}(s)=\left\{\begin{array}{c}(1-\alpha )\frac{{Y}_{{dt}}}{{R}_{{dt}}^{M}},s\in [0,{I}^{* }]\\ 0,s\in ({I}^{* },1]\end{array}\right.$$
(5)
$${l}_{{dt}}(s)=\left\{\begin{array}{c}(1-\alpha )\frac{{Y}_{{dt}}}{{W}_{{dt}}^{L}},s\in ({I}^{* },\theta ]\\ 0,s\in [0,{I}^{* }]\cup (\theta ,1]\end{array}\right.$$
(6)
$${h}_{{dt}}(s)=\left\{\begin{array}{c}(1-\alpha )\frac{{Y}_{{dt}}}{{W}_{{dt}}^{H}},s\in (\theta ,1]\\ 0,s\in [0,\theta ]\end{array}\right.$$
(7)

Under the condition of market clearance, the equilibrium between the supply and demand for robots, low-skilled labor, high-skilled labor, and capital can be expressed as follows:

$${M}_{{dt}}={\int }_{0}^{1}{m}_{{dt}}(s){ds}={\int }_{0}^{{I}^{* }}{m}_{{dt}}(s){ds}=(1-\alpha )\frac{{Y}_{{dt}}}{{R}_{{dt}}^{M}}{I}^{* }$$
(8)
$${L}_{{dt}}={\int }_{0}^{1}{l}_{{dt}}(s){ds}={\int }_{{I}^{* }}^{\theta }{l}_{{dt}}(s){ds}=(1-\alpha )\frac{{Y}_{{dt}}}{{W}_{{dt}}^{L}}(\theta -{I}^{* })$$
(9)
$${H}_{{dt}}={\int }_{0}^{1}{h}_{{dt}}(s){ds}={\int }_{\theta }^{1}{h}_{{dt}}(s){ds}=(1-\alpha )\frac{{Y}_{{dt}}}{{W}_{{dt}}^{H}}(1-\theta )$$
(10)
$${K}_{{dt}}={\int }_{0}^{1}{k}_{{dt}}(s){ds}=\alpha \frac{{Y}_{{dt}}}{{R}_{{dt}}^{K}}$$
(11)

where \({M}_{{it}},{L}_{{it}},{H}_{{it}}\) represent the demand for robots, low-skilled labor, and high-skilled labor, respectively. Taking the logarithm of Eq. (2) on both sides and combining it with Eqs. (8), (9), (10), and (11), the expression can be written as follows:

$$\begin{array}{ll}{ln}\left({Y}_{dt}\right)=\displaystyle{\int }_{0}^{1}{ln}\left({A}_{{dt}}^{s}({x}_{{dt}}({s}))^{1-\alpha }{K}_{{dt}}{({s})}^{\alpha }{ds}\right.\\\qquad\quad =\displaystyle{\int }_{0}^{{I}^{* }}{{ln}}({r}_{{dt}}^{M}(s)({m}_{{dt}}({s}))^{1-\alpha }{K}_{{dt}}({s})^{\alpha }){ds}\\\qquad\quad +\displaystyle{\int }_{{I}^{* }}^{\theta }{ln}({r}_{{dt}}^{L}\left({s}\right)({l}_{{dt}}({s}))^{1-\alpha }{K}_{{dt}}({s})^{\alpha }){ds}\\\qquad\quad +\displaystyle{\int }_{{\theta }^{* }}^{1}{\mathrm{ln}}({r}_{{dt}}^{H}({s}){(h_{{dt}}({s}))}^{1-\alpha }{K}_{{dt}}({s})^{\alpha }){ds}\\\qquad\quad =\left(\displaystyle{\int }_{0}^{{I}^{* }}{r}_{{dt}}^{M}\left(s\right){ds}+\displaystyle{\int }_{{{I}^{* }}}^{\theta }{r}_{{dt}}^{L}\left(s\right){ds}+\displaystyle{\int }_{\theta }^{1}{r}_{{dt}}^{H}\left(s\right){ds}\right)\\\qquad\quad +\left(1-\alpha \right)\left(\right.{I}^{* }{\mathrm{ln}}\left(\frac{{M}_{{dt}}}{{I}^{* }}\right)+\left(\theta -{I}^{* }\right){\mathrm{ln}}\left(\frac{{L}_{{dt}}}{{W}_{{dt}}^{L}}\right)\\\qquad\quad \left.+\left(1-\theta \right){\mathrm{ln}}\left(\frac{{H}_{{dt}}}{{W}_{{dt}}^{H}}\right)\right)+\alpha {ln}({K}_{{dt}})\end{array}$$
(12)

From \(\partial {Y}_{{dt}}/\partial {I}^{* } >\, 0,\partial {I}^{* }/\partial {r}_{{dt}}^{M}(s)\, >\, 0\), we can deduce that the more tasks robots perform, the higher the firm’s output becomes. This indicates that intensifying the substitution of low-skilled labor with robots can boost an economy’s productivity. In perfectly competitive markets where product prices match marginal costs, increased production efficiency leads to lower product prices. Consequently, businesses driven by profit maximization increasingly deploy robots, displacing more skilled labor. Generally, robot applications influence total employment through two main channels: the substitution effect, where robots replace low-skilled workers, thereby reducing their employment opportunities; and the expansion effect, where robots help expand productivity and production scale, thus elevating labor demand. Therefore, the overall impact of robot applications on total employment remains uncertain.

The impact of robot applications on labor transfer

The analysis indicates that robots stimulate the expansion of related industry sectors, thus broadening employment opportunities across the supply chain and resulting in the significant creation of high-quality, skill-intensive jobs. In this context, robots fulfill a dual function: they enable employment opportunities for high-skilled labor and act as substitutes for low-skilled labor. Considering the mobility of the workforce between cities, the total population of city j at time t is represented as follows:

$${{pop}}_{j,t}=\mathop{\sum }\limits_{k=1}^{N}{P}_{j,k,t}{po}{p}_{k,t-1}$$
(13)

Derived from Eq. (13), the net outflow of labor from city j at time t can be expressed as follows:

$${{pop}}_{j,t-1}-{{pop}}_{j,t}=-\mathop{\sum}\nolimits _{-j}{P}_{j,k,t}{po}{p}_{k,t-1}+(1-{P}_{j,j,t}){po}{p}_{j,t-1}$$
(14)
$$\underbrace{\frac{{pop}_{j,t-1}-{pop}_{j,t}}{{pop}_{j,t}-1}}_{Net\,outflow\,of\,labor}= -\underbrace{\sum_{j}P_{j,k,t} \frac{pop_{k,t-1}}{pop_{j,t-1}}}_{Inflow\,from\,other\,cities}+ \underbrace{(1-P_{j,j,t})}_{Total\,outflow\,of\,labor}$$
(15)

Under the condition of other factors remaining constant, taking the partial derivative with respect to the city’s robot density \({M}_{j,t}\) yields:

$$\frac{\partial }{\partial {M}_{{jt}}}\frac{{{pop}}_{j,t-1}-{{pop}}_{j,t}}{{{pop}}_{j,t-1}}=-\mathop{\sum}\nolimits _{-j}\frac{\partial }{\partial {M}_{j,t}}{P}_{j,k,t}\frac{{po}{p}_{k,t-1}}{{po}{p}_{j,t-1}}-\mathop{\sum}\nolimits _{-j}{P}_{j,k,t}\frac{\partial }{\partial {M}_{j,t}}\frac{{po}{p}_{k,t-1}}{{po}{p}_{j,t-1}}+\frac{\partial }{\partial {M}_{j,t}}(1-{P}_{j,j,t})$$
(16)

The equation suggests that the total population of city j in period t depends on the probability of labor migrating to work in city j. This paper characterizes the distribution of workers’ skills using the \(Fr\acute{e}{che}t\) equation (Ahlfeldt et al., 2015; Lagakos and Waugh, 2013) as follows:

$$F({s}_{1},...,{s}_{N})=\exp \left\{-\left(\mathop{\sum }\limits_{k=1}^{N}{s}_{k}^{-\theta }\right)^{1-\rho }\right\}=\exp \left\{-\left(\mathop{\sum }\limits_{k=1}^{N}{s}_{k}^{-\frac{\widetilde{\theta }}{1-\rho }}\right)^{1-\rho }\right\}$$
(17)

where \({s}_{N}\) denotes the skill dispersion within region N, with higher θ values indicating greater disparities in labor skills across regions; ρ signifies the correlation in skill levels between different regions. A higher ρ implies that a worker possesses a high skill level in city j.

Assuming that a laborer’s income from work equals their consumption expenditure, the utility function for worker i choosing to move from city o to city j is expressed as follows:

$${U}_{{ijot}}={\alpha }_{{jt}}{s}_{{ijt}}{q}_{o}(1-{\tau }_{{ijot}}){\varepsilon }_{{jot}}$$
(18)

where \({\alpha }_{{jt}}\) refers to the characteristics of city j, such as living conditions; \({s}_{{ijt}}\) is the skill level of worker i in city j; \({q}_{o}\) denotes the initial skill level); \({\varepsilon }_{{jot}}\) represents unobserved factors. Given the presence of migration costs, based on the “Iceberg Cost” principle, this study assumes that labor migration incurs efficiency losses, denoted as \({\tau }_{{ijt}}\in (\mathrm{0,1})\). The probability of worker i choosing city j is given by:

$$P=\frac{{[{\alpha }_{{jt}}{\varepsilon }_{{jot}}(1-{\tau }_{{ijot}}){q}_{o}]}^{\theta }}{{\sum }_{k=1}^{N}{[{\alpha }_{{kt}}{\varepsilon }_{{kot}}(1-{\tau }_{{ikot}}){q}_{o}]}^{\theta }}$$
(19)

such that \(\partial \theta /\partial \rho\, >\, 0\). As the skill gap among workers across regions widens, their inclination to migrate to these regions diminishes, especially for those with lower skill levels. This study focuses on rural migrants, who generally have education levels below the average. As robot applications increase the demand for higher skills within firms, low-skilled laborers find themselves increasingly marginalized. Based on these arguments, the paper proposes the following hypotheses:

H1: The application of robots increases the migration likelihood among the rural mobile population.

H2: The impact of robots on the migration of rural labor is particularly significant among workers with lower skill levels.

Research and design

Background review

Experiences from developed nations illustrate that the demographic dividend is instrumental in propelling economic growth (Hainmueller and Hisco, 2010; Liu, 2010). During the era of rapid expansion of the labor force between 1980 and 2010, China experienced a substantial surge in its gross domestic product (GDP), averaging an annual growth rate exceeding 10 percent. Population mobility, a central element of the urbanization process, not only enhances the value of human resources but also facilitates the horizontal relocation of industries and the equalization of public services, thereby effectively reducing developmental disparities between regions. The scale of China’s mobile population has continued to rise rapidly, increasing from 221 million in 2000 to 376 million in 2020. However, the total number of migrant workers has been decreasing. In 2020, the total was 286.5 million, a decrease of 5.17 million compared to the previous year, including 169.6 million outbound migrant workers, down by 4.66 million.

As shown in Figs. 1 and 2, the period from 2014 to 2018 witnessed the largest growth in robot installations in China, which has reshaped the labor market landscape. This means that understanding the trends in rural labor mobility is crucial for altering the urban-rural dual structure, guiding the orderly transfer of surplus rural labor, and promoting social harmony and stability (Domini et al., 2021; Chen et al., 2019).

Fig. 1: Industrial robot installations.
figure 1

Source: IFR. The figure demonstrates an overall upward trend in the installations of industrial robots.

Full size image
Fig. 2: Trends in the number of rural migrant workers.
figure 2

Source: National Bureau of Statistics of China. The figure shows a steady expansion in the scale of the rural migrant population.

Full size image

Model setting

Building on the analysis of the aforementioned theoretical models, the regression model formulated to validate the research hypotheses of this study is as follows:

$${Mobilit}{y}_{{ic}ht}={\beta }_{0}+{\beta }_{1}{Robo}{t}_{{ct}}+\sum _{i}{X}_{{it}}+{\mu }_{c}+{\lambda }_{t}+{\delta }_{{ht}}+{\varepsilon }_{{icht}}$$
(20)

where i refers to individual laborers, c refers to the city of employment, h represents the industry sector, t is the year, \({Mobilit}{y}_{{icht}}\) refers to whether an individual laborer i working in industry h in city c during year t intends to relocate again, \({Robo}{t}_{{ct}}\) is the robot density, and \({X}_{{it}}\) accounts for various control variables selected from individual, familial, and economic dimensions. These variables include age, age squared, gender, ethnicity, education level, marital status, family size, presence of family members during relocation, the establishment of a local health record, rental expenditures, duration of migration, local minimum wage, level of economic development, level of secondary employment, and urban human capital. Additionally, the model incorporates fixed effects for cities (\({\mu }_{c}\)), industries-year (\({\delta }_{{ht}}\)) and years (\({\lambda }_{t}\)), with \({\varepsilon }_{{icht}}\) representing the random error term.

In this analysis, the coefficient \({\beta }_{1}\) is of particular interest, as it quantifies the estimated impact of robot applications on the migration tendencies of rural labor. The variables are summarized in Table 1, and logarithmic transformations are applied to continuous variables.

Table 1 Definition of variables.
Full size table

Variable description

Optimizing labor resource allocation is essential for sustaining employment stability and fostering high-quality development. Accordingly, the variables defined in this study are as follows:

Dependent variable: In measuring labor mobility, previous studies have often relied on past transfer behaviors (Ma and Tang, 2020; Bertinelli et al., 2022) and regional migration rates (Leknes et al., 2022; Kirchberger, 2021) for analysis. However, when Giuntella et al. (2022) conducted a study based on CFPS data, they found that the applications of robots had an impact on family fertility, borrowing, and other behaviors. Thus, an analysis based solely on past behavior may introduce certain biases. In this study, the influence of robot applications on the rural labor market was evaluated from the perspective of the likelihood of labor relocating across different regions, utilizing a discrete variable, \({Mobilit}{y}_{icht}\), constructed from responses to the question “Do you plan to reside locally for more than 5 years?”. Responses indicating “do not plan” were coded as 1, while ‘plan’ and ‘undecided’ (indicating a weaker intent to move) were coded as 0. For monthly income, the value comprised the income earned in the previous month (or during the last employment) excluding costs for food and accommodation.

Core explanatory variable: Customs import data and International Federation of Robotics (IFR) data are typically used to establish indicators for urban robot applications. Since customs data no longer provided enterprise names and codes after 2016, IFR data was merged with microdata from the Second National Economic Census in constructing city-level robot density to assess the impact of robot applications on rural labor mobility from 2014 to 2018. Bartik instrumental variables (Goldsmith-Pinkham et al., 2020; Acemoglu and Restrepo, 2020; Wang et al., 2022; Hong et al., 2022) were employed to formulate the urban robot density indicator (\(R{obo}{t}_{ct}\)). Data on robot inventories across various industrial sectors were provided by the International Federation of Robotics (IFR). The 2002 National Economic Industry Classification aligns with the International Standard Industry Classification (ISIC Rev.4), facilitating the acquisition of robot inventories for different industries in China using the corresponding industry codes and data from the IFR. The robot density for each city in each year was calculated by combining the employment figures and robot inventory density for each industry. The formula used for this calculation is as follows:

$$R{obo}{t}_{ct}=\mathop{\sum }\limits_{s=1}^{S}\frac{{labo}{r}_{s,c,t=2008}}{{labo}{r}_{c,t=2008}}\times \frac{R{D}_{{st}}}{{labo}{r}_{s,t=2008}}$$
(21)

Where \({labo}{r}_{s,c,t}\) indicates the number of employees working in industry s within city c during year t, with the manufacturing sector further divided into multiple subsectors; \(R{D}_{{st}}\) refers to the robot density. The robot density for each city for each year is calculated using the provided formula.

Control Variables: Control variables were employed across three dimensions (i.e., individual, familial, and economic factors), which included parameters such as age and its square, gender, ethnicity, education level, marital status, family size, whether family members accompany during relocation, the establishment of local health records, rental expenditures, the duration of migration, the local minimum wage, level of economic development, level of secondary employment, and urban human capital.

Using the median urban robot density as benchmark, the regions were divided into areas depending on robot density. As shown in Table 2, a higher robot density correlates with an increased likelihood of migration among the rural labor force, highlighting the considerable influence of robot applications on local rural labor markets. In areas with relatively high robot density, the average migration probability for the rural labor force (0.216) is higher than that in regions with lower robot density (0.192). Overall, the average migration probability for the rural labor force is 0.204, suggesting that about 20% of the workforce considers further migration. The average educational attainment among rural laborers in high robot-density areas (9.484) surpasses that in low robot-density areas (9.274), indicating higher skills and learning capabilities within regions with greater robot density. This may potentially increase the risk of unemployment for low-skilled workers, thereby influencing migration decisions. The average monthly housing expense for rural laborers in high robot-density areas (474.350) exceeds that in low robot-density regions (380.578). In terms of average minimum wage, low robot-density areas (1331.209) had a much lower mean than high robot-density areas (1614.794). This suggests that economically prosperous regions, given their superior infrastructure and manufacturing capabilities, are more conducive to the development and adoption of robotics technology. In particular, rural migrants generally have lower education levels (9.379), which makes them particularly vulnerable in the labor market, especially in the face of robot applications. Comprehensively analyzing the effects of robotics on the rural labor market, with specific emphasis on the rural migrant population, would have considerable practical importance and hold significant scholarly value.

Table 2 Descriptive statistics.
Full size table

Data sources

The microdata used in this study were obtained from the China Mobile Population Dynamic Surveillance Data (CMDS), collected by the National Health Commission of the People’s Republic of China from 2014 to 2018. These data encompass all 31 provinces (autonomous regions, and municipalities) and the Xinjiang Production and Construction Corps, utilizing a stratified, multi-stage, proportional-to-size sampling method. Sampling locations were chosen randomly in areas with dense migrant populations, targeting individuals at least 15 years old, who had resided in the inflow area for at least a month and did not possess local hukou (household registration) in the region (county or city). Due to its extensive sample coverage and comprehensive survey content, which included basic information, mobility scope, and the income and expenditure status of the migrant population and their family members, the CMDS has become a key database for researching issues related to China’s migrant population (Xu et al., 2023; Meng et al., 2023). The application of robots has restructured labor market demands, presenting significant challenges for rural migrants seeking employment opportunities.

The focus of this study is on individuals aged between 17 and 64 with an agricultural household registration status, whose reasons for migration are employment-related (e.g., seeking work, fulfilling occupational commitments) or engaging in business ventures. Samples from special regions (e.g., autonomous territories and regiments) and those with missing data or involving cross-border migrations were excluded. Monthly income values in the top 1% quantile were truncated. With total valid samples of 150,651, our approach ensures that the focus remains on job transitions influenced by labor market dynamics and not on other non-employment factors. City-level control variables were obtained from various publications, including the China Urban Statistical Yearbook, provincial and municipal Statistical Yearbooks, and Statistical Bulletins. Minimum wage data were aggregated from official announcements by provincial (municipal) Ministries of Human Resources and Social Security and local municipal government sources.

Two primary data sources were utilized in calculating city-level robot density: the International Federation of Robotics (IFR) and microdata from the Second National Economic Census. The IFR database provides data on industrial robot installation and stock from over 50 countries from 1993 to 2019, with the manufacturing sector further segmented into 13 specific industries. In this study, the analysis primarily focused on the secondary sector. The second National Economic Census, which provides data on the scale, structure, and performance of China’s secondary and tertiary industries, encompasses all legal entities, industrial activity units, and individual enterprises within these sectors, providing the foundational data for calculating city-level robot stock density.

In assessing the robustness of urban robot density, customs data from 2014 and 2015 were employed. The customs database provides detailed information on company import and export activities at the product level, and the application of HS product codes enables the identification of the annual quantity and value of imported robots, a methodology well-established in current research (Huang et al., 2022; Fan et al., 2021). The analysis also incorporated data from the 2018 input-output tables. For robustness tests on rural labor migration, our study employed data from three iterations of the China Family Panel Study (CFPS) conducted in 2014, 2016, and 2018. This biennial survey, administered by the China Institute for Social Science Survey at Peking University since 2010, samples urban and rural residents across China.

Empirical analysis

The impact of robot applications on rural labor transfer

The impact of robot applications on rural labor transfer intention

A heteroscedasticity-robust linear probability model (LPM) was utilized to explore the overall impact of robot applications on rural labor migration. Table 3 summarizes the regression results for the various model specifications. Column (1) incorporates the core explanatory variable and includes fixed effects for city, year, and industry-year. Column (2) expands on Column (1) by adding basic characteristics of individuals and families. At the individual level, the variables included age, age squared, ethnicity, gender, and education level, while at the family level, the parameters included marital status, family size, and whether family members are relocating together. Column (3) provides economic attributes like the establishment of health records, monthly rent expenses, and minimum wage levels to the model. Column (4) includes factors related to family migration, such as migration duration, level of economic development, level of secondary employment, and urban human capital, while Column (5) refines the analysis in Column (4) by focusing solely on samples with active employment status. All models applied clustering of standard errors at the city level.

Table 3 Influence of robot applications on rural labor transfer.
Full size table

The regression results show a positive and statistically significant coefficient for robot density at the 5% level. In Column (1), the coefficient for the core explanatory variable, urban robot density, is 0.248, which suggests that on average, a 1% increase in urban robot density is associated with a 0.248% rise in the probability of rural labor migration. This positive and significant relationship persists even with the addition of more control variables. In Column (4), the coefficient for the core explanatory variable was 0.249, suggesting that every 1% increase in urban robot density increased the likelihood of rural labor migration by an average of 0.249%. These findings indicate that increased urban robot application density significantly influences rural labor mobility and that the effects of robot applications (encompassing production expansion and substitution effects) have a substantial impact on the labor market, increasing the likelihood of re-migration among rural workers. Analyzing the coefficients of the control variables, the effect of age on the transfer of rural labor was found to be non-linear, indicating significant heterogeneity in its impact. The probability of continued transfer is higher among male rural workers compared to their female counterparts. Likewise, low-skilled labor exhibited a higher probability of transfer compared to high-skilled labor, with the former primarily involved in routine work and displaying a stronger response to robot impact. Family size was found to negatively impact the likelihood of continued migration within the rural labor force. Larger families usually have more complex considerations to relocate, thereby reducing the propensity to move. Given that medical security supports the social system in ensuring the overall population health, health risks are particularly significant for the floating population. Those with social security benefits are in a better position to manage unforeseen illnesses or accidents, lowering the likelihood of re-migration. In addition, the minimum wage level and monthly rent expenses reflect the economic conditions and living costs in the workers’ current location. Considering the lower employment status of rural migrants, economic factors are critical determinants influencing labor transfers.

Study on the impact of robot applications on urban migration rate

Micro-level individual data from the 2010 population census and the 1% population sample survey in 2015 were employed in constructing the migrant labor migration rate at the prefecture-level city level. This emphasis on labor mobility within prefecture-level cities is due to two key considerations: first, many policies, including those related to land and household registration, are predominantly formulated at this administrative level; second, the interconnectedness and economic interdependence are notably higher within these cities compared to inter-city relations. Based on the regression analysis, a marked decrease was observed in the migration rate of migrant labor in regions where urban robots are widely used, as shown in Table 4. This trend remains consistent even after controlling for population characteristics and employment distribution factors of the original region.

Table 4 Influence of robot applications on migration.
Full size table

Robustness analysis

To ensure the robustness of the findings, several methods were employed to address two potential sources of endogeneity in the research questions. First, reverse causality may exist between robot applications and rural labor migration. A decrease in rural labor mobility may lead to increased labor costs for businesses, prompting a higher adoption rate of robots. Second, omitted explanatory variables (e.g., geographical factors), which can influence the application of robots and the migration of rural labor, may also be present and have to be taken into account.

For the instrumental variable, the density of urban robots from major robot manufacturing countries was used (Goldsmith-Pinkham et al., 2020; Borusyak et al., 2022), given two main considerations. First, the development trend of the robot industry in other developed countries closely mirrors that of China’s robot industry, with trends in principal robot manufacturing nations directly impacting robot usage in China. Also, data from these advanced manufacturing countries can serve as indicators of the level of scientific and technological progress. This suggests that constructing a robot density indicator based on Bartik-IV can mitigate the endogeneity problem to a certain extent.

To minimize potential measurement errors, instrumental variables using data on robots from multiple countries (i.e., Germany, South Korea, the United States, Japan, Sweden, and the United Kingdom) were also employed and the results of the regression analyses are presented in Column (1) of Panel A in Table 5. Further regression analysis using data from Germany, South Korea, and the United States is shown in column (2) of Panel A. Analysis employing data from South Korea and the United States is detailed in column (3) of the same panel. An instrumental variable using the proportion of intermediary values from computer manufacturing and information technology services in total added value was also constructed, as calculated from the 2018 input-output table, with the regression results presented in column (4) of Panel A in Table 5.

Table 5 Influence of robot applications on rural labor transfer -- robustness test.
Full size table

Drawing on the method developed by Ma and Zhu (2022), additional instrumental variables were used for testing using historical data on post and telecommunications from 1984 for each city. The value is calculated by multiplying the number of global mobile network connections from the previous year (\(t-1\)) by the number of landlines per 100 people in 1984. The regression results are shown in column (1) of Panel B, Table 5. In addition, the approach by Fan et al. (2013) was adopted in constructing another instrumental variable, utilizing the historical status of a city as a trading port between 1840 and the end of the Qing dynasty combined with the number of global mobile network connections from the previous year (\(t-1\)). The historical role of a city as a trading port may have long-term effects on local technology adoption, but it is not directly related to current labor migration, thus satisfying the criteria for relevance and exogeneity of the instrumental variables. As shown in Column (2) of Panel B in Table 5, the regression results using the 2SLS instrumental variable approach are generally robust, the coefficients of the core explanatory variables are significantly positive, and the chosen instrumental variables do not suffer from issues related to under-identification or weak instrumental variables.

In constructing the dependent variable, samples responding with ‘not well considered’ were given the value of 0 and were subsequently excluded from the analysis to mitigate measurement errors. As shown in the regression results shown in Column (1) of Panel C in Table 5, the coefficient for the core explanatory variable was 0.263, indicating that a 1% rise in urban robot density increased the average likelihood of rural labor electing to transfer by 0.263%. Given the close interconnection among cities within the same province, the approach used by Nian (2023) was employed for clustering standard errors at a broader level. The regression outcomes, presented in column (2) of Panel C in Table 5, with a modified clustering standard error for the coefficients, exhibit only minor deviations from the earlier findings, indicating consistent robustness in the study’s conclusions.

Considering the substantial sample size and the ability of the linear probability regression framework to effectively incorporate fixed effects and address the issue of omitted variables to a certain extent, the heteroscedasticity robust LPM model was utilized in the analysis. For the robustness check, both Probit and Logit models were employed to confirm the consistency of the conclusions. In the regression results presented in Columns (3) and (4) of Panel C in Table 5, the Logit model had a coefficient of 0.239 for the core explanatory variable. This means that a 1% increase in urban robot density enhances the probability of rural labor choosing to transfer (compared to not transferring) by 0.239%.

To account for potential policy interference, the State Council officially issued the Opinions on Further Promoting the Reform of the Household Registration System on July 30, 2014. This reform aims to reduce the restrictions of the household registration system on labor mobility, encouraging more workers to migrate to different cities for employment opportunities. This policy holds a significant influence on population mobility and urban layout optimization. As shown by the regression results in Column (1) of Panel D in Table 5, the conclusions of this study are consistent and robust.

An alternative estimation method was also utilized to address possible measurement errors in explanatory variables, using the calculation formula structured as follows:

$$R{obo}{t}_{ct}=\mathop{\sum }\limits_{s=1}^{S}\frac{{labo}{r}_{s,c,t=2008}}{{labo}{r}_{c,t=2008}}\times R{D}_{{st}}$$
(22)

The regression results are presented in Column (2) of Panel D in Table 5. To evaluate urban robot applications, robot import data were analyzed from enterprises for the years 2014–2015. The customs database identifies three types of imported robot products: multi-functional industrial robots (HS8 code: 84795010), other industrial robots including end manipulators (HS8 code: 84795090), and automatic handling robots for ICT factories (HS8 code: 84864031). Given the annual operational stock of robots for each enterprise during the specified period, the values can then be aggregated at the city level to calculate the city’s robot density, providing an alternative core explanatory variable. The regression results are detailed in Columns (3) and (4) of Panel D in Table 5. The coefficients for the core explanatory variables in the model were found to be positive and statistically significant, indicating that robot applications increase the likelihood of ongoing migration among the rural labor force.

As an alternative explanatory variable, the change in the type of work performed by the workforce was examined to increase the robustness of the results, the paper uses and improve the analysis of the employment transfer of the rural workforce between agriculture and non-agriculture. Three categories were used: pure labor, part-time, and pure farming. Those not engaged in any agricultural work were given a value of ‘1’, those involved in both agricultural and non-agricultural work were given a value of ‘2’, and those in purely agricultural work were given a value of ‘3’. As shown by the regression results in Table 6, robot application promotes the transfer of rural labor to the agricultural sector and increases the probability of returning to the agricultural sector.

Table 6 Impact of robot applications on labor force job transfers.
Full size table

The effect of robot applications on rural labor transfer -- mechanism analysis

To investigate the mechanisms through which robot applications impact rural labor transfer, this study categorized transfer intentions into passive extrusion and active extrusion based on factors such as employment and insurance status. The influence mechanisms of robot applications on labor migration were analyzed by considering variables such as age, initial skill level, vocational skill level, mobility, and economic disparities in the areas of employment.

Mechanism analysis—transfer intention difference

Drawing on insights from Vadean and Piracha (2010) and Constant and Massey (2002), this paper defines labor migration due to factors such as inability to find satisfactory employment, lack of social security, and household registration issues as passive crowding out. Migration driven by the pursuit of better job opportunities is characterized as active crowding out. From the regression results, presented in Table 7, the coefficient for the actively extruded sample is 0.228, which suggests that a 1% increase in urban robot density would, on average, increase the probability of rural labor choosing to migrate by 0.228%. For passively extruded samples, the coefficient is 0.267, indicating that a 1% rise in urban robot density leads to an average increase of 0.267% in the probability of selective rural labor transfer. This suggests a more significant passive crowding out effect due to robot applications.

Table 7 Influence of robot applications on rural labor transfer—transfer intention difference.
Full size table

The study then explored the relationship between urban robot applications density and the migration of rural labor to cities. The coefficient for the core explanatory variable is −0.109, indicating that each 1% increase in urban robot density decreases the probability of rural labor moving to cities by 0.109%. These findings suggest that robot applications density affects labor migration patterns.

Mechanics analysis—skill differences

The theoretical analysis suggests that robot applications have varied effects on laborers with different skill levels. Without skill upgrading through training and education, workers may encounter increased wage disparities. This could lead to the substitution effect of robots outweighing the expansion effect, potentially displacing low-skilled labor. In this study, these differential impacts were evaluated using three perspectives: age, educational attainment, and occupation.

The regression results for the age-based analysis are presented in Panel A of Table 8. For workers impacted by active extrusion, robot applications significantly affect younger and middle-aged workers. Advancements in robot technology likely generate high-tech jobs, and younger and middle-aged workers, possessing higher technical literacy and adaptability, can more easily transition into these new sectors. Thus, the expansion effect and technological progress associated with robot applications can improve employment prospects for this demographic. In comparison, workers experiencing passive extrusion, particularly those aged 44 and above, show a higher inclination to transfer due to robot applications. The human capital of this group, rooted in work experience and specialized skills, often does not align with changing job demands, making them less competitive in the labor market. Therefore, the effects of robot applications are more substantial for the middle-aged and older workforce.

Table 8 Influence of robot applications on rural labor transfer -- skill difference.
Full size table

The influence of skill level differences was then evaluated on the impact of robot applications on rural labor migration, with the findings presented in Panel B of Table 8. The positive coefficient of the core explanatory variable indicates that robot applications increase the likelihood of rural labor migration. When considering skill level differences, low-skilled labor exhibits a stronger response to robot impacts compared to high-skilled labor. For workers affected by active extrusion, robot technology has a greater impact on high-skilled workers, creating high-paying jobs that demand advanced skills. In comparison, for passively extruded workers, robot technology more significantly impacts low-skilled workers who often cannot meet regional skill requirements, leading them to relocate in search of similar job opportunities. This supports the theoretical hypothesis proposed in this paper.

Lastly, the influence of robot applications on labor migration across different skill levels was explored from an occupational perspective. In the secondary industry, workers are categorized into high-tech and low-tech occupations based on the high-tech industry classification within manufacturing. With increasing urban robot density, dual effects are observed. Many production tasks require high-tech labor for design, operation, and maintenance, while new high-tech job opportunities in areas like robot technology research and development emerge. The regression results in Panel C of Table 8 suggest that the impact of robot applications is more pronounced in low-skilled manufacturing jobs. These jobs are more susceptible to automation, and workers often face skill mismatches, prompting them to relocate to other areas.

Mechanism analysis—liquidity difference

This study explores how robot applications affect rural labor migration, focusing on mobility differences, including the presence of family members during relocation and the scope of mobility. The findings are presented in Table 9. When analyzing the impact of family presence, workers without accompanying family members exhibited a higher inclination to relocate. A 1% increase in urban robot density correlated with a 0.316% increase in the likelihood of these workers considering relocation. Workers relocating with their families, particularly those with children, encounter significant challenges such as securing admissions to educational institutions. As a result, they often seek local employment opportunities, which may compromise their welfare.

Table 9 Influence of robot applications on rural labor transfer -- liquidity difference.
Full size table

In terms of scope of mobility, workers moving across provinces were categorized as highly mobile, while those moving across cities within the same province or counties within the same city were classified as less mobile. Those with high mobility, typically engaged in repetitive blue-collar informal jobs, are more susceptible to replacement by robots as labor costs rise, causing this group to have a higher propensity for migration. The results suggest that a 1% increase in urban robot density is associated with a 0.250% increase in the likelihood of rural labor opting to relocate.

Mechanism analysis -- economic development difference

The differential impact of robot applications on the migration of rural labor was then evaluated by considering the economic development level of employment locations. Based on the “2022 Business Charm Ranking List of Cities,” the first-tier, new first-tier, and second-tier cities were classified as developed cities, third-tier cities were categorized as generally developed cities, and all other cities were classified as less developed cities. The regression results presented in Table 10 indicate that the uneven economic development across China’s regions and the varying density of robot applications have a significant influence on the mobility of rural labor. The impact of robotics is particularly pronounced in developed and more developed areas where rapid technological updates demand a workforce with higher skills to adapt to economic structural changes. The rural migrant population, predominantly educated up to junior and senior high school levels, encounters substantial challenges in meeting the requirements of modernized Industrial systems, often resulting in their displacement. In economically advanced regions, a 1% increase in urban robot density corresponds to a 0.317% increase in the probability of rural labor choosing to relocate.

Table 10 Effect of robot applications on rural labor transfer—economic development difference.
Full size table

Mechanism analysis—the effect of income distribution

The research findings indicate that increased urban robot density significantly enhances labor market mobility and raises the likelihood of rural labor migration. The impact of urban robot applications on the wages and working hours of rural labor was then evaluated, and the regression results are summarized in Table 11.

Table 11 Influence of robot applications on rural labor wages.
Full size table

In Column (1), the results suggest that a 1% rise in urban robot density is associated with an average wage increase of 0.086% for rural labor. This implies that, at the current stage, the positive effects of production expansion from robot applications outweigh any substitution effects. The adoption of robots in China is primarily concentrated on enhancing efficiency and productivity. In Column (2), the findings suggest that a 1% increase in urban robot density corresponds to a 0.093% increase in the average weekly working hours, indicating that companies have enhanced production efficiency by incorporating robot technology. In order to remain competitive and meet the demands of emerging sectors, workers are often required to extend their working hours.

The impact of robot applications on rural labor transfer—further analysis

The analysis reveals that the applications of robots notably increased the probability of rural labor relocation. This study then investigated how different demographic groups are affected by robot applications, with a specific focus on the workers’ gender, marital status, and childbearing status. Given the emphasis enterprises place on efficiency and productivity when adopting robots, the impact of robot applications on the overall technical efficiency was also evaluated from an input-output perspective, in order to provide a more nuanced understanding of the broader economic effects of robot integration in the labor market.

The impact of robot applications on sensitive groups

As shown by the results in Table 12, the increase in the likelihood of relocation caused by robot application was more pronounced among male workers compared to their female counterparts. This discrepancy is likely linked to the prevailing sense of responsibility among women to prioritize family care especially in rural areas, constraining their mobility.

Table 12 The impact of robot applications on sensitive groups.
Full size table

The marital status and number of children of the female respondents were further analyzed, and the results are presented in Columns (3) - (6). The analysis suggests that compared to those who are unmarried, married women have fewer opportunities for skills enhancement and vocational training, limiting their abilities to adapt to technological changes. And in an environment where robotic technologies are becoming more widespread, married women encounter more severe employment challenges and risks.

The effect of robot applications on factor allocation

The findings indicate that at the individual level, robot applications affect labor transfer through both expansion and substitution effects. Increased urban robot density raises the likelihood of rural labor migration. To evaluate whether robot applications can optimize the allocation efficiency of production factors, the DEA-Malmquist index analysis was employed, focusing on technical efficiency from a macro perspective. Urban water and electricity consumption, capital stock (using the perpetual inventory method), and the number of employees at period-end were used as inputs, while real GDP and annual count of inventions were the outputs. The technical efficiency for each city is assumed to start at 1 in 2011, with subsequent years multiplying this figure. However, due to comparability issues, the sample range was restricted to 2014–2018.

As presented in Table 13, the robot impact coefficient is positive and significant at the 5% level, indicating that robot applications enhance technical efficiency. This suggests urban robot density increases production scale and efficiency. Enterprises integrating robots can achieve higher production standards, producing higher-tech, quality-stable products, and channeling capital toward more efficient industries. Therefore, the current utilization of robots has become instrumental in improving the efficiency of factor allocation in the production process.

Table 13 Influence of robot applications on factor allocation.
Full size table

Conclusions and policy implications

Conclusions

Employment is vital for safeguarding and enhancing the welfare of populations. In this study, micro-data from CMDS, IFR, and the Second National Economic Census were employed to examine the effects of robot applications on rural labor migration, as well as the underlying mechanisms and impacts on income distribution and production efficiency. The research reveals that, from a theoretical perspective, robot applications impact labor demand through expansion and substitution effects. Enterprises utilizing robots expand the scale of output and increase the demand for labor by enhancing the automation technology frontier of production tasks and improving overall production efficiency, while simultaneously diminishing the need for traditional labor. Robot applications promote labor mobility between regions when considering relocation costs and increase the probability of rural labor re-migration. Specifically, a 10% increase in robot density was found to correspond to a 2.49% increase in the likelihood of rural labor migration. Urban robot applications reduce the migration rate of urban labor and increase the probability of rural labor returning to the agricultural sector. One notable finding is the pronounced passive crowding-out effect caused by robots, which particularly impacts low-skilled labor and is more pronounced in low-skilled occupations. In terms of mobility, robot applications significantly influence highly mobile groups and may adversely affect the welfare of those migrating with their families. The results also indicate that economically developed regions experience a greater impact from the applications of robots and that their utilization and adoption increase the overall wage levels and extend the workers’ weekly working hours.

Policy implications

Firstly, the significant impact of robot applications on rural labor transfer necessitates a reform of the settlement system. At the core of this reform is the separation of household registration from social welfare benefits, which would promote unrestricted labor mobility and help improve the labor market This would facilitate orderly rural labor migration and dismantle institutional obstacles impeding labor mobility. The reforms should prioritize the cross-regional distribution of public services such as housing, healthcare, and education to promote the equitable sharing of these services across regions, bridge the urban-rural divide, and foster social inclusiveness. Given the potential reduction in welfare for family-accompanied migrant workers resulting from robot applications, targeted government policies are needed, such as relaxing school admission criteria for the children of migrant workers and modifying the healthcare system to more effectively cater to the needs of the elderly in rural migrant communities. Skills training should also be enhanced for these individuals to improve their ability to adapt to the changing labor market dynamics.

Secondly, the pronounced impact of robot applications on low-skilled labor highlights the need for optimizing income distribution among migrant workers. Given the mismatch between job demand and labor supply, the government should actively enhance worker employability and competitiveness by enhancing vocational skills training and consistently investing in skill development, especially for key demographics, to bridge the existing skills gap. The government should also provide comprehensive, ‘one-stop’ services for enterprises seeking skilled labor and for assisting workers in adapting to technological changes and elevating their employment capacity and job quality. To address the displacement effect in low-skilled jobs, the government needs to offer more support and guidance, including unemployment insurance and job transition assistance, to ensure a stable employment landscape.

Thirdly, to address the challenges posed by robot applications to income distribution among migrant workers, the government must actively disseminate employment information, coordinate targeted labor collaborations, and support seamless transitions of workers between industries. This would help mitigate the adverse impacts of robot applications on income distribution. Considering the potential of robots to improve societal production efficiency, China is well-positioned to capitalize on the ‘golden age’ of digital technology. The government should encourage enterprises to adopt robots, artificial intelligence, and other cutting-edge technologies, positioning these technologies at the forefront of the industry’s competition. Such strategic leadership is key to promoting intelligent upgrading within industries and fostering collaborative synergies among enterprises. The government should also champion mechanisms like mergers, reorganizations, or cluster development to horizontally expand the industrial chain, thereby reshaping the distribution pattern and maximizing the complementary effects and technological progress brought about by robot applications.