Childhood Migration Experiences and Entrepreneurial Choices: Evidence from Chinese Internal Migrants

Abstract

Existing research has shown that individuals’ early-life experiences exert a sustained impact on their social life in adulthood. However, there remains a lack of understanding regarding how childhood migration experiences influence entrepreneurial behaviors. Using data from the 2017 China Migrants Dynamic Survey (CMDS), this paper examines the effects of childhood migration experiences on migrants’ entrepreneurial propensity. The findings indicate that childhood migration experiences increase the likelihood of migrants engaging in self-employment or entrepreneurship in China, and this result holds consistent across several robustness checks. The research further reveals that social capital and human capital mediate the relationship between childhood migration experiences and entrepreneurial choices. Additionally, for migrants aged over 35, and those who migrated alone during their first-time migration, the positive effects of childhood migration experiences are more significant. Also, among the three age cohorts of childhood migration, the entrepreneurial effects of migration at ages 7–12 and 13–18 are significantly stronger than those of migration before age 6. This research highlights the long-term impact of childhood migration experiences on shaping individuals’ entrepreneurial choices, which provides theoretical and practical evidence for government policies that promote entrepreneurship.

1. Introduction

In developing countries, differences in industrial structure and levels of urbanization between regions have led to huge internal migration flows. According to data from China’s national census, the country’s floating population surged from 6.57 million in 1982 to 376 million in 2020. Existing research has extensively explored migrants’ employment choices, with a particular focus on their entrepreneurial activities. A consensus exists that migrants are usually risk-tolerant (Halek & Eisenhauer, 2001; Jaeger et al., 2010), a trait identified as a key driver of entrepreneurial decisions (Hvide & Panos, 2014). Additionally, for migrants with limited endowments, entrepreneurship offers a more viable pathway to income improvement compared to wage employment (Fairlie, 2005).

However, there is a critical gap in the existing literature: nearly all discussions have centered on adult migrants who move voluntarily while overlooking a vital group—child migrants who move passively (i.e., second-generation migrants who moved with parents or relatives during childhood and later entered the labor force as adults). This kind of research, which focuses on adult migrants’ entrepreneurial choices, implies an important assumption: the entrepreneurial decisions of migrants are rational responses to the current market environment and their own conditions after arriving at the destination. However, the potential early origins of entrepreneurial propensity have been overlooked. To address this, we introduce the life course theory (Settersten & Mayer, 1997) to provide a key expansion of the existing research. We argue that for the large group of migrant children, their first migration occurs during a critical period for the development of cognitive, non-cognitive, and social skills, and is likely to become a pivotal turning point in their life trajectories. Childhood migration disrupts an individual’s familiar living environment, including familial support networks and community ties. It forces them to adapt to new lives, such as adjusting to a new school and building relationships with peers from diverse cultural or regional backgrounds. One strand of literature examines the impact of childhood migration on short-term life adaptation (Perreira & Ornelas, 2011; Q. Wu & Cebotari, 2018; Schwank, 2024); another strand focuses on the long-term socioeconomic outcomes, such as income levels (Bonikowska & Hou, 2010), social integration (Åslund et al., 2015), and work motivation (L. Chen et al., 2021). Yet, this line of research has rarely extended to long-term economic decisions in adulthood, particularly the impact on entrepreneurship, which is a decision that reflects one’s economic agency. This leaves a gap in our understanding of the long-term impact of early migration experiences among migrant children. Although a relevant study found that childhood migration experiences are associated with migration decisions and even socioeconomic status in later life (Bernard, 2023), it does not address the field of entrepreneurial choices. Therefore, based on this gap, our research shifts the analytical perspective from a static and adulthood-based explanation to a dynamic and life course-based one. We not only focus on an overlooked group but also aim to propose and test a novel theoretical proposition: childhood migration, as a major event in the early life course, can systematically shape the patterns of individuals’ economic behaviors in adulthood through long-term socialization processes, particularly entrepreneurship, which is a high-risk decision. Exploring these aspects is crucial for understanding the long-term economic consequences of population mobility and for providing new insights into existing research on entrepreneurial activities of childhood migrant groups.

Using data from the 2017 China Migrants Dynamic Survey (CMDS), we examined the occupational choices of 95,825 internal migrants from 32 province-level regions in mainland China. We employed a range of empirical methods to figure out the relationship between childhood migration experiences and entrepreneurial choices and found that childhood migration experiences increase the likelihood of migrants engaging in self-employment or entrepreneurship in China. Guided by life course theory, we argue that childhood migration influences adult entrepreneurial choices primarily through two mediating channels: the reshaping of social capital and the development of human capital. Specifically, living in unfamiliar social environments influences the broadening of individual horizons and the development of social integration capabilities (Y. Chen & Wang, 2015), which further affects access to the information, opportunities, and resources necessary for starting a business. Simultaneously, this process of adapting to new environments also shapes the formation of human capital (including educational attainment, physical and mental health), which is strongly associated with entrepreneurial propensity. Additionally, the positive effect of childhood migration experiences exhibits significant heterogeneity across individuals.

This paper makes several contributions by shifting the research focus from voluntary adult migrants to involuntary child migrants. First, it fills the theoretical gap in entrepreneurship research concerning the role of early-life experiences. While prevailing literature, which focuses on voluntary migrants, emphasizes contemporaneous factors such as risk tolerance and resource constraints (e.g., Hunt, 2011; Guerrero et al., 2021; Vandor, 2021), our research incorporates childhood migration experiences into the analytical framework of entrepreneurial behaviors by introducing life course theory as a conceptual foundation. We propose that major early-life events influence entrepreneurial propensity in adulthood, thereby enriching the research perspective on entrepreneurship among migrants. Second, this research identifies the channels through which childhood migration experiences influence entrepreneurial behaviors: namely, the reshaping of social capital and human capital. Even when existing studies examine the mechanisms, such as social or human capital, in migrant entrepreneurship (e.g., Sun & Fong, 2022; Weng & Wang, 2024), they mostly focus on the capital accumulation in adulthood and do not explore how childhood migration experiences are related to these capitals. Through empirical tests of these mechanisms, our research offers a more micro-level understanding of the economic behaviors of migrant children. Finally, our findings offer new insights into policy innovation. Current entrepreneurship policies focus on providing financial support and subsidies for start-ups yet often overlook the cultivation of individual entrepreneurial capacity. Our research suggests that early-life experiences, such as childhood migration, play a critical and lasting role in shaping entrepreneurial intentions and capabilities. Therefore, policies should be shifted from short-term assistance to long-term cultivation: not only addressing the barriers faced by migrant children but also leveraging the potential embedded in their migration experiences through measures such as community support. This will help build a more inclusive entrepreneurial ecosystem, which will not only improve labor market outcomes for vulnerable populations but also promote sustainable economic development.

2. Related Literature and Hypothesis Development

2.1. Social Capital

Social capital is defined as the aggregate of actual or potential resources embedded within durable relationship networks (Bourdieu & Richardson, 1986); alternatively, it refers to the valuable support and knowledge that entrepreneurs acquire through their social networks (Portes, 1998). With the development of the knowledge economy, social capital is recognized as a critical channel for entrepreneurs to access resources (Leyden et al., 2014; Dudley, 2021), particularly in China, a rural society characterized by relationship-based interactions, where formal institutions remain underdeveloped (Freire-Gibb & Nielsen, 2014).

Existing research has examined the impact of social capital on entrepreneurial opportunity identification and entrepreneurial resource acquisition. During the entrepreneurial opportunity identification stage, entrepreneurs engage in activities such as information searching and business opportunity evaluation and assess the feasibility of entrepreneurship. Social capital plays an important role in this process: it influences individuals’ perceptions of the business environment, thereby shaping their entrepreneurial decisions and outcomes (Anderson & Miller, 2003). Individuals with abundant social capital have access to richer information and a greater capacity to identify high-quality entrepreneurial opportunities (De Carolis & Saparito, 2006). At the initial stage of starting a business, social capital is related to obtaining the tangible and intangible resources needed for entrepreneurship (Sorenson, 2018). Drawing on social network theory, accessible social capital can be categorized into two types. The first type refers to close social relationships centered on acquaintances, bonded by ties such as family, neighbors, ethnicity, and religion; these relationships can provide entrepreneurs with informal financing (e.g., private loans and family investments), alleviating entrepreneurial financing constraints (Shao & Sun, 2021). The second type of social capital is formed around shared public interests, hobbies, and so on, which differs from the first type that relies on consanguineous and geographical ties. Such relationships can provide entrepreneurs with the latest business-related information, including insights into market demand and technology applications (Kontinen & Ojala, 2011), thereby expanding their understanding of the business environment (Jack, 2005).

Yet, childhood migration experiences may have uncertain impacts on social capital in adulthood (Myers, 1999). On the one hand, individuals who migrated in childhood are exposed to urban environments at an earlier age, which enhances their ability to build social relationship networks (Carletto & Kilic, 2011) and gain emotional support from family and friends (Fu Keung Wong et al., 2010). These early migrant experiences can positively contribute to their social capital, ultimately increasing their likelihood of entrepreneurial success. On the other hand, early migration can also influence resilience and integration in their subsequent life (Q. Wu et al., 2014). When children migrate, they sever ties with their original communities and may face cultural exclusion when integrating into their new environments, creating barriers to social adaptation (J. Wu & Sun, 2020). Taking China as an example, the structural socioeconomic segregation may disadvantage migrant children, meaning that childhood migration may have negative consequences. For example, the social support system tied to the hukou (household registration identity) system excludes migrant children, which may undermine their sense of belonging (J. Wu & Sun, 2020) and may be detrimental to the accumulation of social capital.

Although existing studies have pointed out the different impacts of childhood migration on social capital, there is still scope for further exploration. On the one hand, the conclusions of current research regarding how childhood migration affects social capital are still inconsistent; on the other hand, from the perspective of the scenarios where social capital functions, existing studies mostly focus on its role in the acquisition of entrepreneurial resources by adult migrants or in the short-term social adaptation of child migrants, while rarely examining the mediating role of social capital between childhood experiences and adult entrepreneurship. In fact, the social skills and network foundations formed during childhood will continuously influence the accumulation of social capital in adulthood, and this long-term accumulated social capital may serve as either an important support or a potential constraint for individuals to engage in entrepreneurial activities in adulthood.

Based on this, we argue that social capital is expected to be a key link in understanding the association between childhood migration experiences and adult entrepreneurial choices. It can not only connect the effect of childhood migration on individuals’ ability to acquire social resources but also directly influence entrepreneurial decision-making and resource integration in adulthood. In light of these insights, we propose our first hypothesis:

Hypothesis 1 (H1). 

Social capital mediates the relationship between childhood migration experiences and entrepreneurial choices.

2.2. Human Capital

Human capital refers to the array of skills, knowledge, experiences, and capabilities that individuals possess, which can be leveraged to generate value in professional settings or through entrepreneurial activities (Martin et al., 2013). The topic of human capital is an important area in entrepreneurship research (Shane & Venkataraman, 2001), as it influences not only entrepreneurial decisions but also entrepreneurial success. First, there is a broad consensus that entrepreneurs with greater human capital, such as specialized skills and business acumen, are better at identifying entrepreneurial opportunities and achieving entrepreneurial success (Bastié et al., 2016). Second, education, particularly higher education and vocational education for entrepreneurship, is positively associated with entrepreneurial intentions (Fayolle & Gailly, 2015; Ni & Ye, 2018), as it enhances individuals’ ability to identify opportunities and willingness to take entrepreneurial risks. Third, work and management experience relevant to entrepreneurship shapes the performance and long-term development of start-ups (J. Liang et al., 2018). Last but not least, as another component of human capital, health status will be directly related to entrepreneurial success (Hatak & Zhou, 2021), improving the general productivity of human capital (Azarnert, 2020).

Existing research has primarily examined how childhood migration experiences affect human capital, exploring both their positive and negative dimensions. In developing countries, children who migrate to regions with more abundant resources often gain access to longer-duration and higher-quality educational opportunities due to their migration (Lu, 2012; Resosudarmo & Suryadarma, 2014). Additionally, compared to children with left-behind experiences in childhood, those with childhood migration experiences tend to have better health outcomes (X. Li et al., 2024), as they migrate with their parents (rather than being left behind) and thus receive more parental companionship and emotional support. However, while most of China’s internal migration reflects families’ aspirations for upward mobility, such environments lay subtle groundwork for tensions in migrant children’s choices between investing in human capital and accumulating wealth (Azarnert, 2012; De Brauw & Giles, 2017). If young people in rural China can earn higher incomes by migrating to cities for low-skilled jobs (compared to engaging in local farming or continuing their studies), they will regard migrant work as a substitute for further education, which is detrimental to the accumulation of human capital. Furthermore, even for children who have chosen to continue their studies, they may face educational inequality and healthcare exclusion as a consequence of migration, primarily due to barriers imposed by the hukou system (Z. Liang & Chen, 2007; G. Ma & Wu, 2019). Specifically, in cities, children holding rural hukou are often limited to enrolling in lower-quality public schools or private schools with higher tuition fees (Xu & Wu, 2022); this further impairs their mental health and academic performance (Fang et al., 2016; G. Ma, 2020). This stands in sharp contrast to the higher-quality public schools accessible to urban children (Xu & Dronkers, 2016). Also, medical subsidies are tied to local hukou status, which increases the difficulty for migrant children to access health services (Y. Huang et al., 2018). Moreover, the migration process disrupts children’s social networks and school continuity (Nusche, 2009), creating challenges for maintaining focus on learning and developing self-management skills.

Existing research on the relationship between childhood migration and human capital presents a notable paradox worthy of attention. On one hand, from the perspective of educational resource acquisition, migration enhances opportunities for children to access higher-quality education. On the other hand, the existence of institutional barriers may weaken or even offset this potential, particularly in the context of internal migration in China, where the structural constraints imposed by the hukou system are highly prominent. Additionally, due to family economic needs, some rural migrant children may choose to enter the urban labor market early to engage in low-skilled work for short-term income, rather than continuing their education; this choice further exacerbates the uncertainty in human capital accumulation. Therefore, there is a need to clarify the ultimate direction of the impact of childhood migration experiences on human capital.

Furthermore, the learning cognitive ability and health status formed during childhood will continuously influence the human capital in adulthood. This long-term accumulated human capital may directly affect an individual’s entrepreneurial decisions. Whether it is the ability to identify entrepreneurial opportunities, the willingness to take entrepreneurial risks, or the professional skills that support entrepreneurial activities, all are closely related to the level of human capital. Based on this, this research argues that human capital is expected to be a key link in clarifying the relationship between childhood migration experiences and adult entrepreneurial choices, and thus proposes the second hypothesis:

Hypothesis 2 (H2). 

Human capital mediates the relationship between childhood migration experiences and entrepreneurial choices.

In summary, as a formative period of human development, childhood represents a critical juncture where migration can reshape social capital and human capital, thereby exerting lasting effects on adults’ economic outcomes, such as occupational choices and entrepreneurial decisions (See Figure 1).

3. Methodology

3.1. Data and Sample

We used data from the China Migrants Dynamic Survey (CMDS), which is administered by the National Health and Family Planning Commission of China (Migrant Population Service Center of NHC, 2023). Targeting individuals aged 15 and above who have resided in their current location for over one month without officially transferring their hukou from their hometown to their place of residence, the CMDS aims primarily to understand the floating population’s living conditions, development status, and migration trends, thereby providing a reference for formulating corresponding policies. It adopts a sampling frame covering all 32 province-level regions and uses the Probability Proportional to Size (PPS) method. This dataset was obtained through a formal agreement, and its use strictly complies with ethical standards and guidelines.

For this study, we utilized the 2017 wave of the CMDS, which covers 439 regions and 169,989 individuals. Because the 2017 wave is the only wave that covers information such as respondents’ employment choices, places of origin and destination, and social integration, it is conducive to explaining the direct and indirect effects of childhood migration experiences on entrepreneurial choices. Given our research focus on examining how childhood migration experiences affect the entrepreneurial choices of the rural migrants in adulthood, we cleaned the data as follows: First, retain individuals aged 18–521; Second, retain those with an agricultural hukou or a hukou status converted from agricultural to residential; Third, exclude individuals who were not in any form of work at the time of the survey2; Fourth, exclude observations with missing values or outliers in core variables. The final dataset consisted of 95,825 observations.

3.2. Variables

3.2.1. Dependent Variable

We measured individuals’ entrepreneurial choices based on their employment status. Here, entrepreneurship is defined as self-employment or starting a business, which differs from wage employment (Cheng & Smyth, 2021; C. Zhao & Li, 2022; Yang & Zhang, 2023). Correspondingly, the variables for entrepreneurial choice and type were constructed based on the question “What is your current employment status?” in the CMDS questionnaire. The entrepreneurship variable was treated as a dummy: a value of 1 was assigned to respondents who reported engaging in self-employment or owning a business, and 0 to all other respondents. With reference to the Global Entrepreneurship Monitor (GEM), entrepreneurship is categorized into two types: self-employment and employer entrepreneurship. Self-employment refers to individuals who work for themselves and primarily rely on their own labor or that of their family members. In contrast, employer entrepreneurship refers to individuals who establish businesses with the potential to grow into large-scale enterprises, thereby creating employment opportunities for a broader population. In this study, self-employment and employer entrepreneurship were operationalized separately: a value of 1 was assigned to each type, respectively, and 0 otherwise.

3.2.2. Independent Variable

Childhood migration experiences are a dummy variable measured by whether the respondent experienced migration before reaching adulthood. According to the UN Convention on the Rights of the Child, a child is classified as anyone under the age of 18. Thus, by calculating each respondent’s age at their first migration3, we defined those whose first migration occurred before age 18 as having childhood migration experiences (assigned a value of 1), with all other respondents assigned a value of 0.

3.2.3. Instrumental Variables

One of our instrumental variables (IVs) is the proportion of childhood migrants from the same county-level place of origin in the CMDS sample, excluding the respondents themselves. This variable aligns with the relevance requirement (Rozelle et al., 1999) because county-level migration culture and information networks (e.g., migrant intermediaries from the same hometown, dissemination of employment information at destinations) significantly influence household migration decisions, ensuring a strong correlation between this proportion and the respondent’s childhood migration experience. The exogeneity of this IV to adult entrepreneurship is ensured by two key factors: it reflects an aggregate-level environmental characteristic (rather than individual-specific traits) that only influences outcomes through migration decisions, not direct interference with entrepreneurial choices (e.g., skill development or market judgment); also, excluding the respondent from calculations eliminates potential autocorrelation with the dependent variable.

Our second instrumental variable is the annual precipitation in the respondent’s county of origin in the year when they first migrated. Using data from the European Centre for Medium-Range Weather Forecasts (ECMWF), we added together the collected monthly rainfall data to obtain the annual precipitation (Muñoz Sabater, 2019). This variable meets the relevance criterion because agricultural income depends heavily on precipitation: droughts or floods reduce crop yields, lowering expected agricultural income and increasing the migration likelihood when income falls below migration costs (Munshi, 2003; Warner et al., 2008). Its exogeneity to adult entrepreneurship is strongly supported by its timing. The precipitation we use corresponds to the specific year when the respondent first migrated, typically a childhood or adolescent year, while the dependent variable reflects economic behavior in later adulthood. There is no plausible causal pathway through which precipitation conditions in a distant childhood year could directly influence an individual’s entrepreneurial choices decades later. Specifically, such early-life climatic conditions do not affect adult access to urban entrepreneurial resources, market judgment, or skill accumulation related to entrepreneurship. Additionally, precipitation is a natural climatic shock, largely independent of subjective household decisions (such as long-term educational investment or entrepreneurial planning) that might confound the relationship between migration and entrepreneurship.

3.2.4. Control Variables

Following previous literature (e.g., Zhang & Acs, 2018; L. Chen et al., 2021), we controlled for a set of individual-level and household-level variables. Additionally, we included province-level dummy variables to capture regional fixed effects (FEs). Given that migration behavior is shaped by both destinations and origins simultaneously, we further incorporated destination-province and origin-province FEs. This specification allows us to better control for both the attributes that attract individuals to migrate into a region and those that drive them to migrate out of their origin, thereby enhancing the accuracy of the model. Definitions of these variables are provided in Appendix A 

Table A1. Definition of main variables.

Variables	Definition
Panel A. Entrepreneurial choices
Entrepreneurship	=1 if the respondent was an employer or self-employed; 0 otherwise
Self-employment	=1 if the respondent was self-employed; 0 otherwise
Employer entrepreneurship	=1 if the respondent was an employer; 0 otherwise
Panel B. Migration experiences in childhood
Childhood migration experience	=1 if the respondent had migration experience in childhood (0–18 years old); 0 otherwise
First migration during preschool	=1 if the respondent migrated for the first time during preschool (0–6 years old); 0 otherwise
First migration during primary school	=1 if the respondent migrated for the first time during primary school (7–12 years old); 0 otherwise
First migration during high school	=1 if the respondent migrated for the first time during high school (13–18 years old); 0 otherwise
Panel C. Instrumental variables
Proportion of migrant children	The proportion of migrant children (excluding the respondents themselves) in the same origin-county
Annual precipitation	The annual precipitation of the origin-county in the year of the respondent’s first migration (unit: meter)
Panel D. Control variables
Individual characteristics
Gender	Male = 1; female = 0
Age	Age of the respondent
Age2	Age squared divided by 100
Education	Years of formal education of the respondent (No formal education = 0; primary education = 6; junior high school = 9; senior high school = 12; college’s degree = 15; bachelor’s degree = 16; master’s or doctoral degree = 19)
Ethnicity	=1 if the respondent was of Han ethnicity; 0 otherwise
CPC membership	=1 if the respondent was member of the Communist Party of China; 0 otherwise
Marital status	=1 if the respondent was married/remarried/cohabiting; 0 otherwise
Health status	=1 if the respondent considered him/herself healthy; 0 otherwise
Chronic disease	=1 if the respondent was clinically diagnosed with high blood pressure and/or type 2 diabetes; 0 otherwise
Household characteristics
Homeownership in destination	=1 if the respondent owned the house that currently lives in; 0 otherwise
Total household size	Total number of household members live together in current residence
Household expenditure	The logarithm of average total monthly household expenditure over the past year (unit: yuan)
Household income	The logarithm of average monthly household income over the past year (unit: yuan)
Siblings	The number of siblings.
Migration characteristics
Scope of current migration	=1 if the respondent migrated from a different province; 0 if the respondent migrated from a different city/county/town within the same province
Total years of migration	The cumulative number of years the respondent has migrated
Number of migration experience	Total number of cities that the respondent has migrated to
Social insurance participation
Health insurance	=1 if the respondent was covered by social health insurance; 0 otherwise
Social Insurance Card	=1 if the respondent has obtained a Social Insurance Card; 0 otherwise
Residence Permit	=1 if the respondent has obtained a Residence Permit/Temporary Residence Permit; 0 otherwise
Regional dummy variable
Origin provinces	32 provinces/municipalities/autonomous regions/Production and Construction Corps
Destination provinces	32 provinces/municipalities/autonomous regions/Production and Construction Corps
Origin cities	331 cities
Destination cities	351 cities
Origin counties	2459 counties
Destination counties	1285 counties

.

3.2.5. Mediating Variables

As analyzed in Section 2, social capital is a mediator through which childhood migrant experiences influence entrepreneurial choices. Based on the sources of social capital, i.e., whether it originates from an individual’s hometown or current place of residence, we categorize social capital into two distinct types. Social capital of the hometown refers to the assets and resources that individuals can access through their social networks rooted in their hometowns. This dimension is operationalized using individuals’ membership in hometown-based organizations, including fellow-townsmen associations and hometown chambers of commerce. In contrast, social capital of the destination includes resources embedded in the new local networks that individuals form in their migration destination; we measure this by assessing participation in local community groups or professional organizations.

Also, human capital constitutes another potential mediator for the relationship. To quantify human capital, we select two relevant indicators: Education and Health status. Further details on the above mediator variables along with original items sourced from the CMDS can be found in Appendix A 

Table A2. Definition of mediator variables.

Variables	Definition	Original Items
Panel A. Social capital
Social capital of the hometown	=1 if respondent has participated in the activities of fellow-townsmen associations, or hometown chambers of commerce; =0 otherwise.	1.Have you participated in the activities of the following organizations in your local area since 2016? Options: A. Fellow-townsmen associations; B. Hometown chambers of commerce; C. Trade unions; D. Alumni associations; E. Volunteer associations; F. Others.
Social capital of the destination	=1 if the respondent has participated in the activities of trade unions, alumni associations or volunteer associations; =0 otherwise.	Same as above.
Panel B. Human capital
Education	Years of formal education of respondents (No formal education = 0; primary education = 6; junior high school = 9; senior high school = 12; college’s degree = 15; bachelor’s degree = 16; master’s or doctoral degree = 19).	1.Level of education. Options: A. No formal education; B. Primary education; C. Junior high school; D. Senior high school; E. College’s degree; F. Bachelor’s degree; G. Master’s or doctoral degree.
Health status	=1 if the respondent considered him/herself “Health” or “Mostly healthy”; 0 otherwise.	1.How would you rate your current health status? Options: A. Health; B. Mostly healthy; C. Unhealthy, but able to take care of myself; D. Unable to take care of myself.

.

3.3. Empirical Design

3.3.1. Baseline Regression

Since the dependent variable is a dummy variable, we used Probit model to examine the relationship between childhood migration experiences and entrepreneurial choices, and established Equation (1) as follows:

$${enp}_i={\alpha }_0+{\alpha }_1{migration}_i+\sum {\alpha }_2{X}_i+{\epsilon }_i$$

(1)

where ${enp}_i$, denotes the dependent variable, which includes three categories of entrepreneurial activity: total entrepreneurship, self-employment, and employer entrepreneurship. The subscript $i$ refers to individual respondents. The independent variable ${migration}_i$ is a dummy variable that captures whether respondent $i$ has migration experiences during childhood (i.e., before the age of 18). $X_i$ comprises individual and household characteristics, as well as dummy variables for the respondent’s current province of residence (destination province) and province of origin (origin province). ${\epsilon }_i$ denotes the random error term.

3.3.2. Mediation Analysis

To further explore whether childhood migration experiences will have an impact on entrepreneurial choices through social capital and human capital (Hypotheses H1 and H2), we conduct mediation analyses. Since traditional mediating effects analysis methods (e.g., the Sobel method and the Bootstrap method) are only applicable to linear models, we adopt the KHB method to estimate the mediating effects (Karlson et al., 2012). This method not only decomposes regression results of nonlinear binary probability models but also allows the inclusion of multiple mediators, enabling a joint test of the mediating effects of social capital and human capital. To better decompose the mediating effects and interpret the regression coefficients, we calculate the average partial effects (APEs). It should be noted that the sample share of employer entrepreneurship is extremely small, accounting for only 5.6% of the total sample (See

Table 1. Descriptive statistics.

Variables	Obs.	Mean	SD	Min.	Max.
Panel A. Dependent variables
Entrepreneurship	95,825	0.415	0.493	0.000	1.000
Self-employment	95,825	0.359	0.480	0.000	1.000
Employer entrepreneurship	95,825	0.056	0.229	0.000	1.000
Panel B. Independent variables
Childhood migration experiences	95,825	0.263	0.440	0.000	1.000
First migration during preschool	95,825	0.010	0.102	0.000	1.000
First migration during primary school	95,825	0.017	0.129	0.000	1.000
First migration during high school	95,825	0.236	0.424	0.000	1.000
Panel C. Instrumental Variables
Proportion of migrant children	95,825	0.263	0.097	0.000	1.000
Annual precipitation	95,825	0.003	0.002	0.000	0.012
Panel D. Control Variables
Gender	95,825	0.568	0.495	0.000	1.000
Age	95,825	34.672	8.504	18.000	52.000
Age2	95,825	12.746	6.066	3.240	27.040
Education	95,825	9.880	3.034	0.000	19.000
Ethnicity	95,825	0.914	0.280	0.000	1.000
CPC membership	95,825	0.034	0.180	0.000	1.000
Marital status	95,825	0.821	0.384	0.000	1.000
Health status	95,825	0.988	0.110	0.000	1.000
Chronic disease	95,825	0.034	0.181	0.000	1.000
Homeownership in destination	95,825	0.228	0.419	0.000	1.000
Total household size	95,825	3.204	1.189	1.000	10.000
Household expenditure	95,825	8.006	0.610	3.912	11.513
Household income	95,825	8.700	0.554	3.912	12.206
Siblings	95,825	0.036	0.214	0.000	4.000
Scope of current migration	95,825	0.502	0.500	0.000	1.000
Total years of migration	95,825	11.156	7.443	0.167	47.333
Number of migration experiences	95,825	2.097	1.947	1.000	88.000
Health insurance	95,825	0.949	0.221	0.000	1.000
Social Insurance Card	95,825	0.513	0.500	0.000	1.000
Residence Permit	95,825	0.689	0.463	0.000	1.000

), which prevents the KHB method from operating stably4. Consequently, our subsequent analysis focuses on the mediating effects of childhood migration experiences on total entrepreneurship.

4. Empirical Results and Analysis

4.1. Summary Statistics

Table 1 provides a descriptive overview of key variables related to migrants. Over 40% of respondents engaged in entrepreneurial activity (as opposed to wage employment), with more than 80% of these entrepreneurs being self-employed and the remaining being employer-entrepreneurs. Additionally, approximately 26% of the sample reported having childhood migration experiences, with the highest frequency of migration occurring during high school. Based on these preliminary descriptive analyses, we will carefully examine the relationship between childhood migration experiences and entrepreneurial choices.

4.2. Baseline Regression Results

Columns (1) and (2) of

Table 2. The impact of childhood migration experiences on entrepreneurial choices.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Entrepreneurship	Entrepreneurship	Self-Employment	Self-Employment	Employer Entrepreneurship	Employer Entrepreneurship
Childhood migration experiences	0.033 *** (0.004)	0.029 *** (0.004)	0.021 *** (0.004)	0.016 *** (0.004)	0.012 *** (0.002)	0.012 *** (0.002)
Individual characteristics	Yes	Yes	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes	Yes	Yes
Dest. Prov. FE	Yes	Yes	Yes	Yes	Yes	Yes
Orig. Prov. FE	No	Yes	No	Yes	No	Yes
Observations	95,825	95,825	95,825	95,825	95,825	95,825
Pseudo R2	0.135	0.165	0.113	0.135	0.121	0.129

Notes: “Dest. Prov. FE” denotes destination-province fixed effects (inflow region fixed effects); “Orig. Prov. FE” denotes origin-province fixed effects (outflow region fixed effects). Columns (1)–(6) provide the marginal effects. The figures in parentheses are Delta-method standard errors. *** denotes the significance level of 1%. All the notes above apply to all subsequent tables.

indicate that childhood migration experiences positively influence entrepreneurial choices. Columns (3)–(6) of Table 2 further explore how childhood migration experiences affect different types of entrepreneurships, revealing that they contribute to both self-employment and employer entrepreneurship, with a stronger effect on the former. This finding, that individuals with childhood migration backgrounds tend to favor self-employment, resonates with J. Chen and Hu (2021), who argued that early migration often pushes individuals toward necessity-driven entrepreneurship due to structural constraints in the labor market. Our research extends this understanding by distinguishing between self-employment and employer entrepreneurship: unlike self-employment, employer entrepreneurship represents a more advanced form of entrepreneurial activity, requiring greater initial capital, higher risk tolerance, and stronger employee management capabilities. Consequently, the smaller coefficient for the impact of childhood migration on employer entrepreneurship may be due to the unique nature of employer entrepreneurship itself, which relies more on post-adulthood resource accumulation than on the early-life adaptability shaped by childhood migration.

4.3. Addressing the Issue of Endogeneity

4.3.1. IV Estimations

If the independent variable is endogenous, the Probit models will fail to generate consistent coefficient estimates. A common solution is to employ the IV-Probit model for Two-Stage Least Squares (2SLS) or Maximum Likelihood Estimation (MLE). However, the IV-Probit model is only suitable for addressing endogeneity when the endogenous variable is continuous. Given that both our independent and dependent variables are dummy variables, we adopted the conditional mixed process (CMP) method proposed by Roodman (2011) with instrumental variables to deal with endogeneity issues.

As shown in Columns (1)–(4) of

Table 3. Results of IV estimations.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Childhood Migration Experiences	Entrepreneurship	Childhood Migration Experiences	Self-Employment	Childhood Migration Experiences	Employer Entrepreneurship
Proportion of migrant children	0.235 *** (0.015)		0.257 *** (0.015)		0.267 *** (0.015)
Annual precipitation	39.491 *** (1.728)		43.307 *** (1.770)		46.814 *** (1.575)
Childhood migration experiences		0.382 *** (0.014)		0.275 *** (0.025)		0.024 ** (0.010)
Individual characteristics	Yes	Yes	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes	Yes	Yes
Dest. Prov. FE	Yes	Yes	Yes	Yes	Yes	Yes
Orig. Prov. FE	Yes	Yes	Yes	Yes	Yes	Yes
N	95,825	95,825	95,825	95,825	95,825	95,825
Wald test of exogeneity	36396.03 ***		27382.49 ***		16788.90 ***
atanhrho_12	−0.844 ***		−0.524 ***		−0.074

Notes: The CMP framework comprises two stages, specified as follows: In the first stage equation, the endogenous explanatory variable is Childhood migration experiences, with the independent variables including instrumental variables and control variables such as Gender, Age, Age², and siblings. In the second stage equation, the outcome variable is Entrepreneurship, and the independent variables consist of the endogenous variable and all the control variables. *** and ** denote the significance levels of 1% and 5%, respectively.

, the atanhrho_12 statistic differs significantly from 0 (p-value = 0). This indicates that the model is endogenous and the results of the joint equation model via the CMP method are more effective than those from single-equation estimation. Although the atanhrho_12 statistic in Columns (5) and (6) of Table 3 is insignificant, the Wald test is significant at the 1% level, confirming that the independent variable is endogenous. Therefore, it is necessary to use the CMP model for instrumental variable estimation5.

Columns (2), (4), and (6) of Table 3 show a notable upward shift in coefficients compared to the baseline regression results. Specifically, the coefficient for total entrepreneurship increases from 0.029 to 0.382, the coefficient for self-employment rises from 0.016 to 0.275, while employer entrepreneurship shows a more modest change (from 0.012 to 0.024). These results likely reflect the correction of downward endogeneity bias in the baseline estimates, particularly pronounced for total entrepreneurship and self-employment. Such bias typically stems from unobserved confounding factors; for instance, family social capital might simultaneously reduce the likelihood of childhood migration (thereby suppressing variation in the independent variable) and enhance entrepreneurial propensity (thereby inflating variation in the dependent variable), which masks the true positive effect in baseline regressions. By isolating the exogenous variation in childhood migration experiences, the IV method effectively mitigates this suppression, revealing real impacts for the first two entrepreneurial categories.

In contrast, the more modest shift in the coefficient for employer entrepreneurship can be explained by two key factors. First, employer entrepreneurship imposes higher entry barriers, such as greater capital requirements and advanced managerial capabilities, making it not easily affected by the confounding mechanisms (e.g., spillover effects of family social capital) that drive bias in estimates for self-employment and total entrepreneurship. Second, the relatively smaller sample size of employer entrepreneurs may limit the precision of endogeneity correction, resulting in more limited adjustments to the coefficient.

These divergent shifts reflect the value of our IV estimations: while baseline models underestimate the entrepreneurial impact of childhood migration, the corrected estimates provide a more accurate reflection of the causal relationship.

4.3.2. Propensity Score Matching

Although the IV approach helps address the impacts caused by unobservable factors, it fails to resolve selection bias related to observable variables. For instance, the instruments do not predict entrepreneurship among non-migrant residents. Thus, we employed the Propensity Score Matching (PSM) method to mitigate potential selection bias. The Nearest-Neighbor (NN) matching method (n = 1) is implemented to obtain estimates.

For more accurate estimation, balance tests are essential to minimize significant differences between the treatment and control groups. As shown in

Table A3. The bias of the mean of the independent variables before and after NN matching.

Variable	Unmatched and Matched	Mean		Bias (%)	Bias Reduction (%)	t-Test		V(T)/
		Treated	Control			t	p Value	V(C)
Gender	Unmatched	0.608	0.554	11.1		15.01	0.000	.
	Matched	0.608	0.622	−3.0	75.7	−3.05	0.002	.
Age	Unmatched	30.035	36.327	−81.1		−106.62	0.000	0.73 *
	Matched	30.036	30.137	−1.3	98.4	−1.62	0.106	1.04 *
Education	Unmatched	9.902	9.872	1.0		1.33	0.185	0.65 *
	Matched	9.902	9.870	1.1	−6.9	1.26	0.206	0.72 *
Ethnicity	Unmatched	0.912	0.915	−0.9		−1.17	0.241	.
	Matched	0.912	0.913	−0.5	45.7	−0.52	0.602	.
CPC membership	Unmatched	0.026	0.036	−5.8		−7.60	0.000	.
	Matched	0.026	0.026	0.1	98.4	0.11	0.911	.
Marital status	Unmatched	0.743	0.848	−26.2		−37.50	0.000	.
	Matched	0.743	0.745	−0.6	97.6	−0.64	0.520	.
Health status	Unmatched	0.992	0.987	4.8		6.21	0.000	.
	Matched	0.992	0.993	−1.1	76.2	−1.51	0.131	.
Chronic disease	Unmatched	0.023	0.038	−8.9		−11.52	0.000	.
	Matched	0.023	0.021	1.0	88.8	1.31	0.190	.
Homeownership in destination	Unmatched	0.220	0.231	−2.6		−3.52	0.000	.
	Matched	0.220	0.216	0.8	68.4	0.93	0.353	.
Total household size	Unmatched	3.267	3.181	7.1		9.92	0.000	1.19 *
	Matched	3.267	3.268	−0.1	98.7	−0.10	0.922	0.85 *
Household expenditure	Unmatched	8.064	7.985	13.0		17.89	0.000	1.08 *
	Matched	8.064	8.070	−0.9	93.0	−1.00	0.317	1.02 *
Household income	Unmatched	8.767	8.676	16.3		22.45	0.000	1.11 *
	Matched	8.767	8.770	−0.6	96.0	−0.70	0.482	1.00 *

* If variance ratio is outside [0.98; 1.03] for Unmatched and [0.98; 1.03] for Matched.

, the bias for all variables was less than 10% after matching6; t-tests revealed no systematic differences in most of the variables between the treatment and control groups, indicating successful matching. We then examined the Pseudo-R², the Mean bias, the B statistic, and the R statistic to verify whether the matchings satisfied the balance hypothesis. As shown in

Table A4. Matching quality indicators.

Sample	Pseudo-R2	LR Statistics (p-Value)	Bias of Mean	Bias of Median	B	R	% of Obs. in the Treated Group with a Suitable Comparison
Unmatched	0.150	16531.40 (0.000)	14.9	8.0	99.7 *	0.83	100
NN matching	0.000	18.99 (0.089)	0.9	0.9	3.9	1.09	60

* If B > 25%, R outside [0.5; 2].

, the value of Pseudo-R² decreases under various matching schemes, implying that there are no disparities in the covariate distributions between respondents with and without childhood migration experiences. Additionally, the B and R statistics confirm that a balanced covariate distribution was achieved after matching across all matching methods7.

Results in Columns (1)–(3) of

Table 4. PSM method and doubly robust estimation.

Effects	(1)	(2)	(3)	(4)	(5)	(6)
	Entrepreneurship	Self-Employment	Employer Entrepreneurship	Entrepreneurship	Self-Employment	Employer Entrepreneurship
	PSM	PSM	PSM	IPWRA	IPWRA	IPWRA
ATT	0.016 *** [0.006]	0.003 [0.006]	0.011 *** [0.003]
ATU	0.025 *** [0.007]	0.015 ** [0.007]	0.015 *** [0.003]
ATE	0.025 *** [0.005]	0.012 ** [0.006]	0.014 *** [0.003]	0.022 *** [0.005]	0.011 ** [0.005]	0.011 *** [0.002]
N	95,825	95,825	95,825	95,825	95,825	95,825

Notes: The outcome variable is Entrepreneurship. Standard errors from 1000 bootstrap replications in brackets. For PSM method, (1) the covariates include Gender, Age, Education, Ethnicity, CPC membership, Marital status, Health status, Chronic disease, Homeownership in destination, Total household size, Household expenditure, and Household income; (2) ATT = Average Treatment Effect on the Treated, ATU =Average Treatment Effect on the Untreated, ATE = Average Treatment Effect. For doubly robust estimation, (1) the control variables of first-stage models include Annual precipitation, Gender, Age, Age², and Siblings; (2) the control variables of second-stage models include all individual and household characteristics variables. *** and ** denote the significance levels of 1% and 5%, respectively.

indicate that the treatment (childhood migration experiences) strongly promotes overall entrepreneurship and employer entrepreneurship, with a weaker effect on self-employment. Specifically, the absolute value of ATT is smaller than that of ATE, and the latter is in turn generally smaller than that of ATU. This implies that if respondents who did not experience childhood migration had done so, their probability of starting a business would increase significantly.

4.3.3. Doubly Robust Estimation

The PSM method has a key limitation: it can only establish an equation that incorporates covariates influencing the outcome variable, but it fails to account for the treatment selection process. Hence, we employed the more robust inverse-probability weights regression adjustment (IPWRA) estimator to address the observable selection bias. A critical advantage of the IPWRA approach is that only one of its two component equations needs to be accurate, a feature known as the “double robustness” property (W. Ma et al., 2018). The model is constructed as follows, where the first equation represents the treatment equation and the second represents the outcome equation:

$$Pr\left({migration}_i\right)={\gamma }_0+{\gamma }_1{Z}_i+{\mu }_i$$

(2)

$$r\left({enp}_i\right)={\delta }_0+{\delta }_1{W}_i+{\vartheta }_i$$

(3)

In Equation (2), $Z_i$ represents a set of factors influencing childhood migration experiences, including Annual precipitation, Gender, Age, Age², and Siblings. In Equation (3), $W_i$ denotes a set of factors influencing entrepreneurship in adulthood, including all control variables related to individual and household characteristics, as well as province FEs. ${\mu }_i$ and ${\vartheta }_i$ are random error terms.

As shown in Columns (4)–(6) of Table 4, the ATE obtained by the IPWRA method differs numerically from that obtained via the PSM method. However, the two sets of results exhibit notable consistency in terms of statistical significance and direction, further proving the reliability of the estimation results.

4.4. Robustness Checks

First, we incorporated additional control variables into the baseline regression. Specifically, we gradually controlled migration-related characteristics and social insurance status (the detailed definitions and measurements of these variables can be found in the Appendix A Table A1), both of which may influence the entrepreneurial choices of the floating population. As shown in Columns (1) and (2) of

Table A5. Other robustness tests for the regression of entrepreneurship on childhood migration experiences.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Entrepreneurship
Childhood migration experience	0.015 *** (0.004)	0.013 *** (0.004)	0.026 *** (0.004)	0.024 *** (0.003)	0.029 *** (0.004)	0.023 ** (0.009)	0.028 *** (0.004)	0.031 *** (0.007)
Individual characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Migration characteristics	Yes	Yes	No	No	No	No	No	No
Social insurance participation	No	Yes	No	No	No	No	No	No
Dest. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Orig. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Dest. City FE	No	No	Yes	No	No	No	No	No
Orig. City FE	No	No	Yes	No	No	No	No	No
Dest. County FE	No	No	No	Yes	No	No	No	No
Orig. County FE	No	No	No	Yes	No	No	No	No
N	95,825	88,258	95,584	93,131	89,312	13,427	65,771	16,598

Notes: Columns (1) to (8) report the marginal effects for probit regressions. The Delta-method standard errors in parentheses. *** and ** denote the significance levels of 1% and 5%, respectively.

,

Table A6. Other robustness tests for the regression of self-employment on childhood migration experiences.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Self-Employment
Childhood migration experience	0.010 *** (0.004)	0.015 *** (0.004)	0.015 *** (0.003)	0.020 *** (0.003)	0.016 *** (0.004)	0.012 (0.010)	0.015 *** (0.004)	0.019 *** (0.007)
Individual characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Migration characteristics	Yes	Yes	No	No	No	No	No	No
Social insurance participation	No	Yes	No	No	No	No	No	No
Dest. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Orig. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Dest. City FE	No	No	Yes	No	No	No	No	No
Orig. City FE	No	No	Yes	No	No	No	No	No
Dest. County FE	No	No	No	Yes	No	No	No	No
Orig. County FE	No	No	No	Yes	No	No	No	No
N	95,825	88,258	95,708	93,131	89,312	13,435	65,771	16,598

Notes: Columns (1) to (8) report the marginal effects for probit regressions. The Delta-method standard errors in parentheses. *** denotes the significance levels of 1%.

and

Table A7. Other robustness tests for the regression of employer entrepreneurship on childhood migration experiences.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Employer Entrepreneurship
Childhood migration experience	0.004 *** (0.002)	0.011 *** (0.002)	0.011 *** (0.002)	0.013 *** (0.002)	0.012 *** (0.002)	0.009 * (0.005)	0.012 *** (0.002)	0.011 *** (0.003)
Individual characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Migration characteristics	Yes	Yes	No	No	No	No	No	No
Social insurance participation	No	Yes	No	No	No	No	No	No
Dest. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Orig. Prov. FE	Yes	Yes	No	No	Yes	Yes	Yes	Yes
Dest. City FE	No	No	Yes	No	No	No	No	No
Orig. City FE	No	No	Yes	No	No	No	No	No
Dest. County FE	No	No	No	Yes	No	No	No	No
Orig. County FE	No	No	No	Yes	No	No	No	No
N	95,808	88,241	93,968	76,705	89,295	13,414	65,771	16,562

Notes: Columns (1) to (8) report the marginal effects for probit regressions. The Delta-method standard errors in parentheses. *** and * denote the significance levels of 1% and 10%, respectively.

, the coefficients of Childhood migration experiences remain significantly positive.

Second, we controlled for regional fixed effects in dimensions different from the baseline regression (i.e., city and county levels). Columns (3) of Table A5, Table A6 and Table A7 reveal that after sequentially controlling for city-level destination FE and origin FE, the estimated coefficients remain consistent with those from the baseline regression. Columns (4) of Table A5, Table A6 and Table A7 confirm that our core findings remain robust even after accounting for county-level FEs.

Third, we restricted the sample to individuals who migrated for employment purposes (excluding all other migration motives), using the multiple-choice question “What’s your reason for this migration” from the 2017 CMDS. As presented in Column (5) of Table A5 and Table A6, the marginal effect of childhood migration experiences in this restricted sample aligns closely with that of the baseline regression.

Fourth, to examine whether our results are interfered with by the economic development differences between the places of destination and the places of origin, we categorize the respondents into three groups: childhood migration to economically worse-off provinces compared to province of origin, childhood migration to moderately better-off provinces compared to province of origin, and migration to much better-off provinces compared to province of origin8. We then re-estimate the models for each group, and the results (see Columns (6)–(8) of Table A5, Table A6 and Table A7) show that our findings are basically robust. Childhood migration significantly promotes both total entrepreneurship and employer entrepreneurship across all three groups. For self-employment, however, the effect is only significant in the latter two groups (moderately better-off and much better-off provinces) and insignificant in the first group (worse-off provinces). This heterogeneity can be attributed to the following: In worse-off destination provinces, the lack of entrepreneurial resources (e.g., market scale, policy support, and industrial agglomeration) may offset the potential positive impact of migration on self-employment. In contrast, moderately or much better-off provinces provide sufficient resources to translate childhood migration experiences into tangible entrepreneurial behavior, even for necessity-driven entrepreneurial motives.

4.5. Mediating Effects Analysis

Table 5. Mediation effect of social capital and human capital.

Panel A: Decomposition using the APE Method
Effect type	Coefficient		95% conf. interval
Total effect	0.048 *** {0.004}		0.041 to 0.055
Direct effect	0.031 *** {0.004}		0.024 to 0.038
Indirect effect	0.018
Panel B: Summary of confounding
Variable	Confounding ratio	Confounding percentage		Distributional Sensitivity
Childhood migration experiences	1.573	36.43		0.999
Panel C: Effects of mediator variables
Mediator Variables	Coefficient	Percentage of indirect effect		Percentage contribution of mediators
Social capital of the hometown	0.00038 {0.00011}	2.20		0.80
Social capital of the destination	0.00026 {0.00029}	1.50		0.55
Education	0.01689 {0.00063}	96.46		35.14
Health status	−0.00003 {0.00005}	−0.16		−0.06

Notes: Standard errors in braces (not available under the APE method). The coefficients of the mediator variables are reported to five decimal places because some values were close to zero. *** denotes the significance level of 1%.

reports the results of the mediating effect analysis using the KHB model. As presented in Panel A of Table 5, childhood migration experiences have a significant positive effect on entrepreneurial choices (coefficient = 0.048, p < 0.001). After incorporating the mediators, the direct effect remains significant but diminishes (coefficient = 0.031, p < 0.001), with an estimated indirect effect of 0.018. Panel B of Table 5 further reveals that the total effect of childhood migration experiences is 1.573 times the magnitude of its direct effect, with 36.43% of the total effect of childhood migration experiences transmitted through the mediators. This confirms that social capital and human capital mediate the relationship between childhood migration and entrepreneurial choices, and the hypotheses H1 and H2 could been validated.

Panel C of Table 5 reports the effects of each mediator, as follows:

• Social capital of the hometown exhibits an indirect effect of 0.00038, accounting for merely 0.80% of the total effect. From a life course perspective, childhood migration may enable individuals to retain some social connections and capital from their family or hometown, which could shape adult entrepreneurship through the continuity of such resources. However, this effect is negligible in our analysis. One possible explanation is that migration weakens original ties (Z. Huang et al., 2023), offsetting their positive role and resulting in only a weak positive trend. A more critical reason may lie in the structural limitations of such a type of social capital: migrant populations (and their families) often come from regions with relatively poor rural regions, where the local resources supporting entrepreneurship, such as access to funding, market information, and industrial networks, are inherently scarce (J. Zhao & Li, 2021). Even if individuals’ social ties to their hometowns are not completely severed, the resources embedded in these relationships may be insufficient to meet the demands of entrepreneurship. In other words, for the migrant children in this study, the social capital derived from their hometowns that they possess and can access may be more focused on providing basic support for life adaptation and emotional needs, rather than supporting high-risk entrepreneurial activities. This explains why the mediating pathway shows only a weak trend in our model.
• Social capital of the destination has an indirect effect coefficient of 0.00026, contributing 0.55% to the total effect. While life course theory suggests that early migration experiences are expected to promote the restructuring of social networks and the accumulation of social capital in the destination, our results provide no substantive empirical evidence for this pathway. The absence of a significant effect may be attributed to a dilemma faced by individuals who experienced childhood migration. First, as migrant children or adolescents, their limited social integration capacity (a consequence of early-life migration) hinders the effective establishment of high-quality new networks. Second, structural constraints in the destination further exacerbate this issue: migrant groups (especially those who moved in childhood) often face subtle social exclusion or cultural barriers (Hermansen, 2017; Xia & Ma, 2020). Even if they form new networks, these ties tend to be limited to other migrant groups with similar socioeconomic backgrounds, lacking the diversity and resource richness required to facilitate entrepreneurship.
• Education has an indirect effect coefficient of 0.01689, accounting for 35.14% of the total effect, making it the most core mediator. Among all stages of life, the return on human capital investment in childhood is the highest, and thus the impact of early migration on individuals’ educational trajectories will have long-term effects on their subsequent development. Unlike some studies that emphasize educational exclusion faced by migrant children (Xia & Ma, 2020), we do not observe the negative effects of migration on education here. This discrepancy may originate from the characteristics of our sample. As shown in Table 1, the proportion of respondents who first migrated during high school is 23.6%, which is much higher than that of migrant children in other age groups. For this group, the barriers to urban compulsory education, often tied to hukou, had minimal influence, which may explain why we did not find a negative impact of migration on education. Moreover, our findings align with those of Sieg et al. (2023), which demonstrate that implementing hukou system reforms helps improve the educational quality of migrant children in third-tier cities, thereby increasing their college enrollment rates. Furthermore, for individuals with childhood migration experiences, greater educational attainment plays a positive role in facilitating their entrepreneurial decisions, which validates the positive effect of education on entrepreneurship (Millan et al., 2014), as education enhances individuals’ ability to process information, evaluate risks, and access entrepreneurial resources critical for initiating and sustaining business activities.
• Health status has an indirect effect coefficient of −0.00003, accounting for just −0.06% of the total effect. Early-life migration may have an underlying long-term impact on an individual’s health status due to sudden changes in living environments, which could influence their subsequent economic decision-making (including entrepreneurship). Our results indicate that while childhood migration might indirectly reduce the probability of starting a business by lowering health levels, this is not a primary pathway.

4.6. Heterogeneity Analysis

4.6.1. Gender Grouped Regression

Existing studies have found that men are generally more likely to engage in entrepreneurship than women (Strawser et al., 2021; Crane, 2022). This disparity results from a combination of economic, cultural, and other dimensional factors. From an economic perspective, investors (or finance institutions) often lack confidence in women entrepreneurs’ debt repayment capabilities or perceive greater uncertainty in female-led ventures, making it difficult for women to access the necessary capital for starting a business (Coronel-Pangol et al., 2024; Ackah et al., 2024). Meanwhile, compared with men, women face disadvantages in accessing support systems such as social networks and legal assistance (Aidis et al., 2007), further reinforcing the perception of underperformance in female entrepreneurship (Amoroso & Link, 2018). From a cultural perspective, a stereotypical gender norm has been emphasizing that men should be breadwinners (Lee, 2022); conversely, women should be caregivers in the household, meaning that they are expected to bear greater household responsibilities (Jayachandran, 2021). This makes it difficult for women to fully commit to entrepreneurial activities (De Clercq et al., 2022), particularly in countries that have long been influenced by agricultural culture (Elkafrawi & Refai, 2022; Rahman et al., 2023).

Although we infer that there are gender differences in entrepreneurial activities, the results in Columns (1) and (2) of

Table 6. Heterogeneity Effects.

Entrepreneurship
Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Male	Female	First Migration Before Age 6	First Migration Between Age 7–12	First Migration Between Age 13–18	Age ≥ 35	Age < 35	Alone at the Time of the First-Time Migration	First-Time Migration with Family/Friends
Childhood migration experiences	0.029 ** (0.005)	0.030 *** (0.006)	0.005 (0.015)	0.043 *** (0.011)	0.024 *** (0.004)	0.029 *** (0.007)	0.018 *** (0.004)	0.058 *** (0.005)	0.00004 *** (0.005)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Dest. Prov. FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Orig. Prov. FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
p value of the coefficient difference	0.434		0.046	0.360	0.000	0.018		0.000
N	54,447	41,374	95,825	95,825	95,825	44,004	51,813	40,390	55,429
Pseudo R2	0.153	0.185	0.164	0.164	0.165	0.121	0.186	0.180	0.155

Notes: Delta-method standard errors in parentheses. Columns (1), (2), (6), (7), (8), and (9) report the results of Fisher’s Permutation test for coefficient differences between respective groups. Columns (3), (4), and (5) present the results of the SUEST test for coefficient differences between groups. Specifically, Column (3) reports the SUEST results for “First Migration Before Age 6” group and the “First Migration Between Age 7–12” group, Column (4) for the “First Migration Between Age 7–12” group and the “First Migration Between Age 13–18” group, and Column (5) for the “first migration before age 6” group and the “First Migration Between Age 13–18” group. *** and ** denote the significance levels of 1% and 5%, respectively.

show that childhood migration experiences have no significant gender difference in their impact on adulthood entrepreneurship (p-value of the coefficient difference = 0.434), regardless of how entrepreneurship is defined. It may be because the main form of entrepreneurship among China’s migrants is subsistence-oriented self-employment entrepreneurship (such as food stalls and domestic services). These entrepreneurial activities place lower demands on factors prone to gender disparities, such as physical strength, risk tolerance, and social capital, thereby making the impact of childhood migration experiences more consistent between men and women. In addition, the phenomenon of population mobility has changed the traditional gender division of labor in rural areas, where the breadwinner-caregiver division of family roles is highly prevalent (X. Zhao et al., 2023); an increasing number of women are getting involved in work and, like men, they are striving to earn higher incomes (Liu, 2020; Z. Chen & Barcus, 2024). This change in the gender structure of the labor force has made it so that the impact of childhood migration on entrepreneurial choices is no longer constrained by traditional gender norms.

4.6.2. Childhood Migration Stage Grouped Regression

To further identify at what age childhood migration experiences have a more significant impact on entrepreneurial choices, we draw on the life course perspective and replace the dummy Childhood migration experiences with three age-specific migration dummies, categorizing the childhood migration experience into three critical developmental stages: 0–6 years (early childhood), 7–12 years (middle childhood), and 13–18 years (adolescence). This categorization is theoretically and empirically grounded. First, 0–6 years is a period of rapid neurocognitive and emotional development (Bai et al., 2022), where the external environment lays the foundation for the formation of preschool children’s temperament and personality (Son & Morrison, 2010; Cohodes et al., 2021). Migration during this stage may disrupt the environmental stability (Gong & Rao, 2023), though it could also foster early adaptability to new environments. Second, 7–12 years aligns with primary school education, a key phase for building social skills, academic identity, and local social networks. Migrant primary school students often face obstacles in access to public education in addition to acculturation stress (Z. Liang et al., 2020; Ren & Jiang, 2021), which is conducive to mental health and human capital accumulation. Third, 13–18 years is a period of identity formation and autonomy development, where adolescents begin to form career aspirations and internalize social norms around work and success (Napolitano et al., 2021; Kaland, 2021). During this period of time, migrants can either choose to receive education or engage in the labor market (Azarnert, 2012). This binary choice shapes how migration at ages 13–18 influences their future entrepreneurship, with two contrasting effects. On one hand, if they engage in the labor market, the urban occupational experiences they gain can build informal practical skills and adaptive mindsets, laying the groundwork for later entrepreneurial attempts. On the other hand, prioritizing labor market participation over education may weaken their accumulation of foundational human capital that supports long-term entrepreneurial success.

We first examine the SUEST (Seemingly Unrelated Estimation and Testing) for coefficient differences between groups. As shown in Columns (3)–(5) of Table 6, the p-value of the coefficient difference test indicates that there is no significant difference in the impact on entrepreneurship between migrating during primary school (7–12 years) and high school (13–18 years). In contrast, the coefficient difference tests for the other two pairs (i.e., migration before age 6 vs. migration between age 7–12, and migration before age 6 vs. migration between age 13–18) are statistically significant (p < 0.05). This also implies that the mechanism through which preschool migration affects entrepreneurship is significantly different from that of school-age migration. Turning to the regression coefficients, the coefficient of First migration before age 6 is insignificant (p > 0.1), while the coefficients of First migration between age 7–12 and First migration between age 13–18 both show significant and positive effects. Notably, primary school age emerges as a critical dividing line for migrant children: those migrating before age 6 often face barriers such as limited access to public schools in destination areas, frequent school transfers, and inadequate support for social and academic integration. Thereby, the coefficient is small and not significant. In contrast, children migrating at ages 7–12 and 13–18, who have already entered primary school, exhibit significant entrepreneurial effects, although the difference between these two groups is not significant.

4.6.3. Age Grouped Regression

It is widely recognized that entrepreneurship, being a long-term and high-risk investment endeavor, is more likely to be pursued by younger individuals with greater risk-bearing capacity (Arenius & Minniti, 2005). Empirical research utilizing individual data has consistently demonstrated an inverse U-shaped relationship between age and the propensity to start a business, indicating that the likelihood of initiating new firms peaks among younger and middle-aged individuals (Evans & Leighton, 1990; Levesque & Minniti, 2006; Minola et al., 2016; Zhang & Acs, 2018), with successful entrepreneurs in most countries concentrated among young and middle-aged individuals aged 25–35 (Azoulay et al., 2020). However, some studies suggest that older individuals, having accumulated greater managerial experience and social capital, may exhibit higher entrepreneurial propensity (e.g., J. Liang et al., 2018). Therefore, we conduct grouped regressions across age to investigate whether age heterogeneity in entrepreneurial choices exists.

We infer that the entrepreneurial choices of China’s migrant population may differ depending on whether their age is above or below 35 and examine the heterogeneity effect using a cutoff at 35 years of age. The reason is that age discrimination is prevalent in China’s labor market, specifically referring to the difficulties faced by individuals over 35 years old, including declining occupational competitiveness, reduced employment opportunities, and concentrated outbreaks of life pressure (Yuan, 2023). This kind of disadvantage brought about by age may affect an individual’s career choice and entrepreneurial decision. In addition, this age-based grouping approach is also commonly adopted in existing literature (Knight et al., 2011; Athukorala & Wei, 2018).

As shown in Columns (6) and (7) of Table 6, the positive effect of childhood migration experiences on entrepreneurial probability is significantly stronger among older individuals. This can be attributed to the fact that older individuals have typically accumulated more professional experience, resources, and social connections, enabling them to better leverage childhood migration experience to support business success. Also, their accumulated work experience, paired with early migration backgrounds, fosters a broader worldview and access to more entrepreneurial opportunities (Miralles et al., 2016). Furthermore, for people over 35 years old, starting a business is a promising way to earn a higher income to support a family, compared with being employed. In contrast, younger individuals are generally in the early stages of their careers, where they face higher uncertainty and risk. For this group, childhood migration may lead to insufficient resource accumulation or limited disposable wealth, undermining their confidence and capacity to launch entrepreneurial projects (Hincapié, 2020).

4.6.4. Status of First-Time Migration Grouped Regression

We further group migrants by whether they moved alone or with family/friends in their first migration. This grouping is based on the idea that migration companionship shapes social networks, resource access, and psychological support (Wong & Leung, 2008; Fu Keung Wong et al., 2010), which in turn affect outcomes like entrepreneurial choices. However, the effect of companionship during the migration is uncertain. W. Q. Li et al. (2022) empirically found that companionship of a significant other reduces an individual’s motivation to make new friends, whereas Tang et al. (2024) shows that the companionship from parents increases the well-being of migrant children, which further improves their academic/career development and performance. When it comes to an individual’s first migration (at a relatively younger age), the role of companionship may be even greater. Thus, we examine the heterogeneity effect across the status of first-time migration.

Columns (8) and (9) of Table 6 reveal that individuals who migrated alone at their first-time migration are significantly more likely to engage in entrepreneurship, highlighting the role of individual independence traits. Migrating alone at first migration might foster an individual’s self-reliance, proactive network building, and motivation to seize opportunities, thus having a stronger impact on entrepreneurial choices, while migrating with others may lead to relying on existing social ties, resulting in a weaker effect.

5. Conclusions

This research examines the impact of childhood migration experiences on entrepreneurial choices, using data from the 2017 CMDS. The findings indicate that childhood migration experiences play an important role in influencing individuals’ entrepreneurial choices, and the results remain robust across several robustness checks. Among all the channels through which childhood migrant experiences influence entrepreneurial choices, education explains the majority of the indirect effects. This emphasizes the significant impact of childhood migration experiences on an individual’s educational attainment, which will largely influence entrepreneurial choices. Heterogeneity analysis further reveals that individuals aged 35 and above, as well as those who migrated alone during their first-time migration, exhibit a significantly higher probability of engaging in entrepreneurship. In addition, the entrepreneurial effects of migration at school age are significantly stronger than those of migration at preschool. This research enriches the occupational choice literature, particularly regarding the link between childhood experiences and the entrepreneurial choices of the floating population. It also emphasizes that entrepreneurial choices are not only influenced by an individual’s current financial situation, personality traits, etc., but are also related to their earlier childhood experiences. Overall, this work provides a direction for future research on entrepreneurial behaviors and a new perspective for boosting residents’ entrepreneurship in the context of China’s large-scale population mobility.

Our research has policy implications. First, given that childhood experiences exert substantial impacts on shaping individuals’ future life trajectories, cities should enhance their inclusivity by accelerating the provision of basic public services to the floating population and their children (Gabrielli & Impicciatore, 2022). For instance, the children of the floating population should be included in the coverage of urban community public services, and the responsible entities for the connection of public services such as education, medical care, and culture should be clearly defined to ensure the accuracy and timeliness of service supply. Second, to stimulate the development of employer entrepreneurship, the government should take further measures to support start-ups, for example, by establishing digital platforms dedicated to enhancing entrepreneurial capabilities. Third, targeted support should be extended to groups with higher entrepreneurial potential identified in this research, individuals aged 35 and above and those who migrated alone for the first time. For instance, local governments could design entrepreneurial training programs and provide microloan preferential policies for these groups of migrants, addressing their potential resource constraints and maximizing their entrepreneurial willingness. Fourth, since education accounts for the majority of the indirect effect of childhood migration experiences on entrepreneurial choices, migrant-receiving areas should optimize educational resources for migrant children via institutional reforms. Specifically, governments should advance hukou system reforms to break the link between hukou status and access to public education resources, ensuring migrant children enjoy equal enrollment opportunities in high-quality public schools as local children. The design of the current education system should be explicitly upgraded to allocate educational resources based on the permanent resident population of the city rather than the registered population. The certification materials required for school enrollment should be simplified, and unreasonable rules should be abolished or strictly prohibited, such as requirements that both parents must reside locally. In areas where school enrollment quotas are tight, a points-based enrollment system may be implemented (Bo & Wang, 2025), but its fairness and transparency must be ensured. Additionally, efforts should be made to encourage the top-tier primary and junior high schools to run branch campuses. These high-quality public schools are expected to extend their premium resources, including excellent teaching staff, mature management experience, and well-designed curriculum systems, to newly established schools and underperforming ones in areas with concentrated migrant populations. This measure will help narrow the gap in educational quality between schools and promote the balanced distribution of educational resources within the region. Meanwhile, more attention should be paid to migrant children’s physical and mental health. For instance, by building a community-based social support networks to provide migrant children with sufficient emotional and academic support, laying a solid foundation of social capital and human capital for their potential entrepreneurial endeavors in the future.

This paper also has several limitations. First, some rural-born migrants have returned to their hometowns to start businesses, yet this group was not included in our sample. Existing research shows that the social capital and human capital accumulated in cities increases these migrants’ probability of engaging in entrepreneurship in their hometowns (Ding et al., 2025). Thus, the impact of childhood migration experiences on entrepreneurship may be underestimated in our analysis. Future studies could focus on tracking the fate of migrants returning to their hometowns, to fully capture the long-term impact of early migration experiences on entrepreneurial trajectories across spatial contexts. In addition, by linking migrants’ social and professional assimilation experiences in destinations to their subsequent entrepreneurial behaviors in hometowns, future studies can clarify how specific assimilation outcomes shape the success of their return entrepreneurship. In other words, the degree of urban assimilation is likely a key determinant shaping whether and where they ultimately choose to become entrepreneurs. Second, while our definition of entrepreneurship follows established practices in Chinese entrepreneurship research, it may be broader than the traditional conception of the term. Future studies could advance this work in two key ways: One approach is to focus on surveying entrepreneurs who employ a certain number of employees; this could be operationalized by re-estimating the models after excluding solo vendors, or by applying an earnings or licensing threshold. Another strategy involves precisely defining self-employed individuals excluding cases where this status is imposed by employers, for instance, when they wish to avoid formal labor contracts. These adjustments would help uncover the true effect of childhood migration on entrepreneurship, thereby addressing the current limitations of overly broad classification. Third, certain family-of-origin characteristics may influence both childhood migration decisions and entrepreneurial choices, but we were unable to account for these factors due to the survey design constraints. While we attempted to mitigate this issue by employing a two-stage model with exogenous instrumental variables, this aspect could be further improved in future research by using data from long-term longitudinal surveys. Fourth, our analysis may not fully capture the psychological mechanisms through which childhood migration experiences affect entrepreneurial behaviors, for example, mental health and non-cognitive ability. In future studies on this topic, psychologically relevant survey items could be designed to better examine how childhood migration experiences shape personal traits and thereby influence entrepreneurial choices.